Patent Application
20160089612
1. Field of the Invention
The invention is in the technical field of electronic toys. More specifically, the invention pertains to mobile toys such as electronic cars and model railroads.
2. Description of Related Art
Many electronic toys are controlled by a human operator. Such examples include radio and remote controlled cars and model trains that are controlled through a handheld device.
These kinds of toys have little or no ability to sense and interact intelligently and flexibly with their environment. Also, they do not have the ability to adjust their behavior in response to the actions of other toys. Further, many toys are physically constrained to slot or track systems and are therefore restricted in their motion.
The invention is a toy system that includes a drivable surface comprised of a plurality of segments, e.g., without limitation, a straight segment, an intersection segment, a left-curve segment, a right-curve segment, a left-turn segment, a right-turn segment, and/or any other suitable and/or desirable segment that can be envisioned. Each segment includes markings which encode locations on the segment and which encode a location of the segment on the drivable surface. The toy system also includes at least one toy vehicle (or mobile agent). The toy vehicle (or mobile agent) can take on any suitable and/or desirable form, such as, without limitation, a vehicle (e.g., a car, a truck, an ambulance, etc), an animal, or any other desired form. The toy vehicle includes at least one motor for imparting motive force to the toy vehicle, an imaging system operative for taking images of the markings, a vehicle wireless transceiver, and a microcontroller operatively coupled to the motor, the imaging system, and the vehicle wireless transceiver. The microcontroller is operative for controlling, via the motor of the toy vehicle, detailed movement of the toy vehicle on the drivable surface based on images taken of the markings of the drivable surface by the imaging system. Lastly, the toy vehicle includes a basestation comprising a controller and a basestation wireless transceiver operatively coupled to the controller. The controller is operative for determining via wireless communication from each vehicle wireless transceiver to the basestation wireless transceiver a current location of the toy vehicle on the drivable surface based on images taken of the markings of the drivable surface by the imaging system of the toy vehicle. The controller stores a virtual representation of the drivable surface and determines based on said virtual representation and the current location of each toy vehicle on the drivable surface an action to be taken by the toy vehicle on the drivable surface, such as, without limitation: vehicle speed, vehicle acceleration, vehicle direction/heading, the state of at least one light of the vehicle, and/or a sound output by an audio speaker of the vehicle. Lastly, the controller communicates to the microcontroller of each toy vehicle the action to be taken by the toy vehicle on the drivable surface via wireless communication from the basestation wireless transceiver to the vehicle wireless transceiver.
The microcontroller of each toy vehicle can be responsive to the action communicated by the controller for controlling the detailed movement of the toy vehicle on the drivable surface based on images taken of the markings of the drivable surface by the imaging system to cause the toy vehicle to move toward a future location on the drivable surface. The detailed movement of the toy vehicle comprises the microcontroller's detailed implementation of the action communicated by the controller, which action comprises one or more acts to be performed by the toy vehicle to move to the future location. More specifically, the future location resides in the controller as a static or dynamic location where the controller desires the toy vehicle to move. The action communicated by the controller to the microprocessor comprises one or more actions to be performed by the toy vehicle in furtherance of the overall goal or plan of movement of the toy vehicle to the future location. Lastly, the detailed movement of the toy vehicle comprises, for each action to be performed by the toy vehicle, one or more steps to be taken by the toy vehicle in furtherance of the action.
As can be seen, the future location, the one or more actions to be performed by the toy vehicle, and the detailed movement/steps to be taken by the toy vehicle represent a distributed command hierarchy, with the future location stored at the controller being at the top of the hierarchy, the one or more actions to be performed communicated by the controller to the microprocessor in the middle of the hierarchy, and the detailed movement/steps to be taken by the toy vehicle being at the bottom of the hierarchy. Each successively lower level of this distributed control hierarchy comprises increasingly more detailed instructions/commands in furtherance of a higher level command. For example, without limitation, the microcontroller may need to implement a number of steps in fulfillment of an action (e.g., change lanes to the left) communicated to the microcontroller by the basestation. Similarly, the basestation may need to implement a number of actions in fulfillment of the overall goal or plan of movement of the toy vehicle to the future location.
The toy system can also include a plurality of toy vehicles. The controller can be operative for controlling the interaction of the plurality of toy vehicles on the drivable surface in a coordinated manner with each other via wireless communication from the basestation wireless transceiver to the vehicle wireless transceivers of the plurality of toy vehicles.
The controller can be operative for controlling at least one of the following of at least one of the plurality of toy vehicles: a velocity or acceleration of the toy vehicle; a set, e.g., row, of markings (driving lane) the toy vehicle follows on the drivable surface; a changing of the toy vehicle from one set of markings (driving lane) on the drivable surface to another set of markings (driving lane) on the drivable surface; a direction the toy vehicle takes at an intersection of the drivable surface; the toy vehicle leading, following, or passing another toy vehicle on the drivable surface; or activation or deactivation of a light, an audio speaker, or both of a toy vehicle. The controller can also be operative for updating control software (e.g., without limitation, firmware) of the vehicle that controls the operation of the vehicle microprocessor.
The toy system can also include a remote control in communication with the basestation, wherein the basestation is responsive to commands issued by the remote control for controlling at least one of the following via the basestation: which one of a plurality of toy vehicles is responsive to the commands issued by the remote control; a velocity or acceleration of a toy vehicle responsive to commands issued by the remote control; a changing of a toy vehicle responsive to commands issued by the remote control from one set of markings (driving lane) on the drivable surface to another set of markings (driving lane) on the drivable surface; a direction a toy vehicle responsive to commands issued by the remote control takes at an intersection of the drivable surface; a toy vehicle responsive to commands issued by the remote control leading, following, or passing another toy vehicle on the drivable surface; or activation or deactivation of a light, an audio speaker, or both of a toy vehicle responsive to commands issued by the remote control.
The controller can be operative for controlling, either in response to or in the absence of a response to movement of a remote controlled vehicle, at least one of the following of each toy vehicle not under the control of the remote control: a velocity or acceleration of the toy vehicle; a set of markings (driving lane) the toy vehicle follows on the drivable surface; a changing of the toy vehicle from one set of markings (driving lane) on the drivable surface to another set of markings (driving lane) on the drivable surface; a direction the toy vehicle takes at an intersection of the drivable surface; the toy vehicle leading, following, or passing another toy vehicle on the drivable surface; or activation or deactivation of a light, an audio speaker, or both of a toy vehicle.
The drivable surface can include at least one multi-state device (e.g., a traffic light. a railroad crossing gate, a draw bridge, a trap on a road piece, a garage door, etc.) responsive to the controller for changing from a one state to another state.
The imaging system can include a light source outputting light toward the markings and an imaging sensor for detecting light from the light source reflected from the markings.
A layer can cover the markings of at least one segment. The layer can be transparent to light output by the vehicle's imaging system but opaque at human visible light wavelengths. The markings can be visible or invisible at frequencies detectable by humans.
The controller can be responsive to the current location of the toy vehicle on the drivable surface and the virtual representation of the drivable surface for causing a display to display the following: a virtual image of the drivable surface and a virtual image of at least one toy vehicle and its position, velocity, or both on the virtual image of the drivable surface.
The drivable surface can be comprised of a plurality of discrete segments operatively coupled together.
The invention is also a method of controlling movement of one or more self-propelled toy vehicles (or mobile agents) on a drivable surface that includes markings which define one or more paths of toy vehicle travel on the drivable surface and which encode locations on the drivable surface, wherein each toy vehicle includes an imaging system for acquiring images of the markings. Each toy vehicle (or mobile agent) can take on any suitable and/or desirable form, such as, without limitation, a vehicle (e.g., a car, a truck, an ambulance, etc) an animal, or any other desired form. The method comprises (a) while traveling on the drivable surface, a toy vehicle acquiring an image of a portion of the markings of the drivable surface via the toy vehicle's imaging system; (b) responsive to the image acquired in step (a), the toy vehicle controlling its movement on the drivable surface; (c) the toy vehicle wirelessly communicating to a basestation data regarding a location on the drivable surface where the portion of the markings in step (a) were acquired; (d) the basestation responsive to the data communicated in step (c) for updating a position of the toy vehicle in a virtual representation of the drivable surface; (e) the basestation determining an action to be taken by the toy vehicle on the drivable surface based on the data regarding the location on the drivable surface of the portion of the markings acquired in step (a); and (f) the basestation wirelessly communicating to the toy vehicle said action to be taken by the toy vehicle on the drivable surface.
The method can also include repeating step (a)-(f) at least one time. Step (b) can include the toy vehicle being responsive to the action communicated in step (f) for controlling its movement on the drivable surface.
Step (b) can include the action communicated in step (f) causing the toy vehicle to change from traveling on a first path defined by a first set of markings to a second travel path defined by a second set of markings, whereupon the action communicated in step (f) includes said second travel path.
Step (b) can also include the toy vehicle controlling its velocity, its acceleration, its steering direction, a state of one or more of its lights, whether a vehicle audio replication device outputs sound.
The method can also include the basestation determining the virtual representation of the drivable surface from one of the following: a definition file accessible to the basestation; exploration of the physical layout of the drivable surface by one or more toy vehicles acting under the control of the basestation and communicating information regarding the physical layout of the drivable surface to the basestation; or a bus system of the drivable surface which is comprised of a plurality of segments, wherein each segment includes a bus segment and a microcontroller that communicates with the basestation and with the microcontroller of each adjacent connected segment via the bus segment.
Step (a) can include acquiring the image of the markings via an overlayer that is transparent to the toy vehicle's imaging system but which is opaque at human visible light wavelengths.
The method can include repeating steps (a)-(f) for each of a plurality of toy vehicles, wherein: step (e) can also include the basestation determining for each toy vehicle a unique action to be taken by the toy vehicle; and step (f) can also include the basestation wirelessly communicating to each toy vehicle the unique action to be taken by said toy vehicle on the drivable surface in a manner whereupon the plurality of toy vehicles move in a coordinated manner on the road.
The method can also include the basestation receiving a command for the toy vehicle from a remote control, wherein step (e) can also include the basestation determining the action to be taken by the toy vehicle on the drivable surface based on the command received from the remote control.
The present invention will be described with reference to the accompanying figures.
The present invention is a system of toy vehicles that can drive autonomously through an environment without being physically constrained to a slot or track. The vehicles use specially designed sensors that allow them to determine their position in an environment. This position information is processed by software (e.g., without limitation, artificial intelligence (AI) software) running on a separate computer or basestation. Operating under the control of this software, the basestation decides on actions for the vehicles and sends them high level controls. Whereas previous vehicles require entirely human control, the software can control the vehicles and can command them to execute complex actions. This allows the vehicles to interact and respond to the actions of other vehicles as well as other objects in the environment.
While each vehicle can be controlled autonomously via the software, hybrid control is also allowed. This allows users to take control of one or more specific vehicles from the basestation. The basestation continues to track the behavior of all of the vehicles and adjusts the behavior of the vehicles not under user control in response to the user-controlled vehicle(s). Users can decide which vehicle(s) is/are controlled by the basestation and which are controlled by the users.
The vehicles drive on a drivable surface which is a series of road pieces (e.g., straight, left turn, right turn, intersection) that are physically connected together. They can drive forward and backwards, and can also turn freely. This is fundamentally different from other toys that utilize connected drivable pieces and are physically constrained to a slot or track. Further, the vehicles use sensing and control technologies to determine the location of the drivable lanes on the road. This allows the vehicles in the system to interact and execute complex behaviors as described above. The vehicles can also choose to leave a road and drive to another part of the environment. This is another significant difference from toys that utilize a slot or track system.
Using encoding technology, position information is embedded into each road piece. As a vehicle travels over a road piece, it emits light that is reflected by the road piece and the reflected light encodes information about the vehicle's position. The vehicle's sensor detects this encoded light and a microcontroller of the vehicle can use it to decode position and other information. This process can be hidden from users by using emitted and reflected light that is outside the normal human visible spectrum (e.g., infrared or ultra-violet).
The system has two primary modes of operation, racing and non-racing. In racing mode, the road pieces are designed like a race track, and many vehicles can travel in close physical proximity to each other as they travel around the drivable surface. In non-racing mode, the road pieces are designed like standard streets and highways such as those found in typical urban driving. Here the lanes are appropriately spaced apart and vehicles can choose to follow traffic rules and deal in appropriate ways with other vehicles and road situations. These two modes can be combined together to build drivable surfaces that have both racing and non-racing sections.
With reference to
1 Road Pieces:
With continuing reference to
1.1 Piece Location and Type Identification:
With reference to
These markings 12 can encode information such as the identity of the type of road piece 6 the vehicle 2 is currently driving on (e.g., straight, intersection, etc.), unique locations on that road piece 6, and a line 10 to suggest an optimal position for the vehicle 2 if it desires to stay within its lane. Herein, this line will be referred to as a center-line 10 but it is important to realize that the vehicle 2 is in no way required or constrained to follow this line. In the example shown in
Each piece ID 14 encodes a unique piece type within the network and remains constant throughout a road piece 6. Each location ID 16 encodes a unique location on that particular road piece 6 and counts up, desirably, from 0. The example segment of
Since piece and location IDs 14, 16 are assumed to be of the same bit-length, stop-bars normally appear on each side of center-line 10 simultaneously. Thus, one can therefore represent special information from each road piece 6 at locations that have a stop-bar only on one side and a normal marking on the other. Such techniques can be used to encode the direction of the road piece 6 (left, straight, or right) in the first marking row of each road piece to suggest a steering direction to vehicle control software (discussed hereinafter).
With reference to
Desirably, some of the information encoded in the markings 12 is interpreted directly by the vehicle 2, while other portions are relayed back to basestation 22. Basestation 22 interprets the codes parsed by the vehicle 2 from these markings 12 and has an internal (virtual) representation of the drivable surface and each possible road piece 6 type, allowing it to identify each vehicle's 2 exact position in the drivable surface and consider this position and the positions of other vehicles in the drivable surface in its commanded behaviors of the vehicle 2. This also allows future expansion or custom-built road pieces 6 with only small software updates to basestation 22 rather than having to also update each vehicle 2.
A software tool used to generate road piece 6 marking schemes can be used to generate road piece surfaces while customizing a variety of parameters including but not limited to bar widths, spacing between bars, spacing between marking rows, the number of marking rows per full location or piece ID, the number of lanes in each direction of traffic, road piece length, road piece curvature, etc. A final option allows the addition of a checksum bar on each marking row to serve as an error checking tool by encoding the parity of the remaining bars.
Such an encoding scheme is not possible within an intersection (shown in
The same road piece structure can be used for both street driving and racing versions of the system, except that for the racing version, road pieces are single direction and include tightly-spaced lanes (see e.g.,
Markings 12 serve several purposes. First, markings 12 allow vehicles 2 to identify the road piece 6 type that they are on during drivable surface initialization (described hereinafter). Next, markings 12 allow the encoding of various parameters, such as the curvature direction of the road piece 6 upon entering a new road piece 6, thus enabling vehicles 2 to better handle control-related challenges. Additionally, markings 12 provide position estimates at sufficiently fine resolutions to allow basestation 22 to create high-level plans and interactive behaviors for vehicles 2. Finally, each vehicle 2 is able to accurately maintain a heading within a lane 20 using the center-line 10 and estimate its speed and acceleration using the periods of time for which markings are visible or not visible since the precise lengths of the bars and spaces between them are known.
1.2 Structure Identification:
Desirably, basestation 22 knows the exact structure of the drivable surface. Since a user is free to reconfigure the road pieces 6 at any time, there are a variety of techniques that enable basestation 22 to identify the structure of the drivable surface. This structure is defined by a series of road piece 6 connections. For example, knowing all road piece 6 types and that, for example, connection point “2” of road piece “5” connects to connection point “1” of road piece “7” informs basestation 22 the exact relative position and orientation of the two road pieces and, if one of the road pieces is already fixed, this anchors the other's global position in the drivable surface. Since connection information is often redundant, the structure can be uniquely identified without exhaustively identifying all connection pairs. Once all road pieces are anchored to the global coordinate frame, basestation 22 has complete knowledge about the structure of the drivable surface.
In the following section, three techniques are described how the exact structure of a drivable surface can be determined by basestation 22.
1.2.1 Reading from File:
The easiest way to identify the drivable surface is to read its definition from a user-defined definition file. This is a simple and effective method, especially for development purposes.
1.2.2 Instant Electronic Identification:
With reference to
Using bus system 24, basestation 22 can determine which road pieces 6 are in the network. Besides the bus itself, each road piece 6 needs only one digital input/output line per connector. An input/output line allows a microcontroller in each road piece 6 to choose whether the line is used as an input line to read a signal or an output line to generate a signal. For example, a straight road piece has two connectors 26 (one on each end), and each connector 26 has one digital input/output line. A four-way intersection piece has four connectors 26 and therefore four digital input/output lines, one per connection point.
In this setup, a single digital I/O line per connection and a minimalistic bus system are enough to allow basestation 22 to exactly determine the structure of the connected road pieces 6. This is achieved by basestation 22 causing the digital I/O lines on each road piece to sequentially turn-on, while all the other I/O lines are configured as inputs and listen for signals. Then, basestation 22 interrogates each road piece 6 if and where it saw an ON signal. The following describes the method used to determine the road structure using the example in
1.2.2.1 Initialization:
With reference to
1.2.2.1.1 Determine Structure:
These steps are repeated for each unknown connection until the drivable surface structure is fully identified. This method is extremely efficient (the computational complexity scales linearly with the number of road pieces 6 in the network) so basestation 22 can react quickly to any changes in the structure. The bus system is interchangeable with possible alternatives including SPI, CAN, I2C, One-Wire, or a wireless network technology, as long as bus 24 is capable of determining unique IDs from each node and connection point.
1.2.3 Exploring Vehicles:
With reference to
A method to accomplish this task with one vehicle 2 is to track unexplored road piece 6 connection points as this drivable surface is constructed and repeatedly plan paths through the closest unexplored exit until no unexplored exits remain. An example of this method is illustrated in
In
As shown in
As shown in
Multiple vehicles 2 can more quickly identify the drivable surface representation when doing this processing simultaneously but must take into account any uncertainty in their locations in order to prevent collisions. For example, two intersection road pieces 6 in the system may have the same type so the vehicles 2 must be sure that if there is uncertainty about which piece they are on, they are performing actions that under any possible scenario will not cause them to collide during exploration.
While this approach results in cheaper road pieces 6 since we reuse existing vehicles 2 and imaging systems 3 for drivable surface identification, it requires a full drivable surface exploration by vehicles 2 each time a change is made.
1.2.4 Ability to Print Custom Networks:
Optionally, software can be provided that runs on an optional general purpose computer and which gives a user the ability to design drivable surfaces having custom road pieces. While standard road pieces 6 will include the most common types, such as without limitation, straight sections, turns, intersections etc., some users may want to custom-design non-standard road pieces 6. This software will allow the user to do so, or even design entire sophisticated drivable surfaces. For example, a user could design a single, sharp 45 degree turn or a large scale racing track with extra wide roads and pit stop exits.
Using this software, the user can request that each non-standard road piece 6 or even an entire drivable surface be printed on, for example, without limitation, paper or transparency. The user can then attach this printout to his preferred surface.
The software can also provide a definition file for the custom designed network which can be uploaded to the user's basestation 22, so that the basestation 22 understands the custom road piece(s) 22 and/or drivable surface and can plan appropriate actions for the vehicles 2.
2 Vehicles:
With reference to
Each vehicle 2 can be fully controlled by basestation 22 or through hybrid control between a user via a remote control 132 and basestation 22. If a user controls a vehicle 2, he can choose to have the vehicle 2/basestation 22 handle low level controls such as steering, staying within lanes, etc., allowing the user to interact with the system at a higher level through commands such as changing speed, turning directions, honking, etc. This is useful since vehicles 2 are small and can move fast, making it difficult for a human to control the steering.
2.1 Vehicle System Components:
Vehicles 2 described herein are different from remote controlled toy cars available today. To allow for the above-described behaviors, vehicles 2 include various robotics and sensor technologies. Each vehicle 2 includes five main system components described in the following sections and illustrated in
2.1.1 Microcontroller:
A microcontroller 40 is the main computer on each vehicle 2. It performs all the control functions necessary to allow vehicle 2 to drive, sense, and communicate with basestation 22 and monitor its current state (like position in the drivable surface, speed, battery voltage, etc.). Desirably, microcontroller 40 is low cost, consumes little power but is powerful enough to intelligently deal with large amounts of sensor data, communications requirements, and perform high speed steering and speed control. Microcontroller 40 includes a variety of peripheral devices such as timers, PWM (pulse width modulated) outputs, A/D converters, UARTS, general purpose I/O pins, etc. One example of a suitable microcontroller 40 is the LPC210x with ARM7 core from NXP.
2.1.2 Wireless Network Radio:
Each vehicle 2 includes a wireless network radio 42 (i.e., a wireless radio transceiver) operating under the control of microcontroller 40 to facilitate communication between microcontroller 40 and basestation 22. Potentially, many vehicles may be driving on the drivable surface simultaneously and basestation 22 needs to communicate with all of them, regardless of whether they are controlled by users, by basestation 22, or both. This means that vehicles 2, traffic lights 8, basestation 22, and remote controls 132 for user interaction can be part of a wireless network which can handle multiple (potentially hundreds of) nodes. The network topology can be set up as a star network, where each vehicle 2, remote control 132, etc. can communicate only with basestation 22, which then communicates with vehicles 22, remote control 132, etc. The second possibility is to choose a mesh network topology, where nodes can communicate directly with other nodes.
The star network version is simpler and requires less code space on microcontroller 40 but still fulfills all the requirements needed for the system. Examples of suitable wireless network technologies include ZigBee (IEEE/802.15.4), WiFi, Bluetooth (depending on the capabilities of future versions), etc. Specifically, ZigBee (IEEE/802.15.4) or related derivatives like SimpliciTI from Texas Instruments offer the desired functionality like data rate, low power consumption, small footprint and low component cost.
2.1.3 Imaging System:
The imaging system 3 of vehicle 2 allows the vehicle 2 to determine its location in the drivable surface. A 1D/2D CMOS imaging sensor 46 (shown in
As described above, road pieces 6 include a structured pattern of optical markings 12 that are desirably visible only in the near infrared spectrum (NIR), in the IR (infrared) spectrum or in the UV (ultra violet) spectrum, and are completely invisible to the human eye. This is achieved by a very specific combination of IR, NIR, or UV blocking ink and a matching IR, NIR, or UV light source. Invisible markings are desired since the markings are not visible to the user, making the appearance of road pieces 6 closer match that of real roads. However, without changes in the hardware, the same system works with visible ink as well (such as black), allowing users to print their own road pieces on a standard printer without having to buy special cartridges. The 1D/2D CMOS imaging sensor in the vehicle, together with an LED light source 44 (shown in
For example, a 1D linear pixel array TSL3301 from TAOS INC or a MLX90255BC from Melexis can be used as the image sensor 46. The image of the surface of a road piece 16 can be focused with a SELFOC lens array 48 and illuminated by an NIR LED light source 44 emitting light for example at 790 nm. The pattern of markings 12 on the road pieces 6 would in that case be printed with an ink or dye which absorbs NIR light. The peak absorption frequency is approximately the same wavelength as that at which the LED light source 44 is emitting light, 790 nm in this example. A marking therefore appears black to the imaging system 3 and the surface without a marking appears white (see
The combination of LED light source 44 with peak emitting frequency approximately equal to the peak absorption frequency of the ink completely eliminates the necessity of a light filter if light from outside light sources can be shielded by the vehicle's body. This is the case with the vehicle 2 design shown in
2.1.3.1 Location within the Vehicle:
With reference to
As described herein, imaging system 3 is used to gather sensor information allowing vehicle 2 to steer and keep a desired position on a road piece 6. The distance d1 between the image sensor 46 of imaging system 3 and steering axis 50 is important because it significantly influences a vehicle's 2 capability to steer.
Initially it may seem that imaging system 3 should be located as close as possible to steering axis 50 with the optimal position being at steering point 52 between drive wheels 54 (as shown in
Unfortunately, vehicle 2 is subject to a certain amount of steering noise caused by limited control over motor behavior, slip, backlash, variable friction and other factors. By positioning imaging system 3 forward from the steering axis (as shown in
The optimal position for the imaging system 3 is a tradeoff that must be considered when designing the vehicle 2. If it is positioned too close to the steering axis 50 then the problem mentioned above will occur. On the other hand, if it is positioned too far forward then tiny steering errors will result in large translations for the imaging system 3, possibly adding difficulty to the marking-parsing process. A compromise between the two extremes will allow the vehicle 2 to more easily maintain its heading on a road piece while still enabling consistent parsing of road markings.
Given the size of the vehicles 2 envisioned herein, a large d1 is desirable, for example d1=25 mm. This allows vehicles 2 to be designed without wheel encoders (sensors to measure angular position and velocity of the wheels) or similar sensors. Usually, sensors like wheel encoders are used to allow cars, robots etc. to steer precisely. Without such sensors, the ability of controlling wheel speed is limited, which causes large steering errors. However, in the vehicle 2 design described herein, the vehicle 2 compensates for such steering errors with a large d1, catching a potential steering error early enough to compensate for such error.
As described herein, this causes the imaging system 3 to change position when the vehicle steers, potentially missing parts of underlying road markings. To account for this position change caused by steering errors, the road markings 12 are designed to be smaller than the imaging systems sensor area, leaving large enough margins on either side.
2.1.4 Motor Drive Unit
With reference to
In another embodiment, the motor drive unit is a single micro electric motor driving at least one rear wheel of the vehicle. Microcontroller 40 generates a PWM signal that is amplified by a motor driver (not shown) to output the necessary power for said motor. In this embodiment, the front wheels of the vehicle can both turn like a real vehicle, e.g., Ackermann steering.
The imaging system 3 described above is also used to estimate a speed of vehicle 2 and steering angle and therefore allows for closed loop speed and steering control. In addition, microcontroller 40 can use a counter-E.M.F. signal from one or both motors 58 to estimate the speed of vehicle 2 at a higher frequency without the need for wheel encoders, but with less accuracy than with the imaging system 3.
2.1.5 Secondary Input/Output:
Each vehicle 2 can also include a secondary input/output system which includes components which are not critical for the core operation of the vehicle 2 but which adds functionality to the vehicle 2 to allow for more realistic performance.
Each vehicle 2 includes a small battery 64 which powers the vehicle 2. The preferred battery type is Lithium Polymer, but other battery technologies can also be used provided the batteries are small. The vehicle uses an A/D converter in series with a voltage divider to enable microcontroller 40 to measure the voltage of the battery. This information is forwarded to basestation 22, which then plans accordingly. When battery 64 is at very low voltages, microcontroller 40 can react immediately and stop the operation of vehicle 2 if necessary. Battery 64 is connected to the bottom of vehicle 2 to supply outside-accessible charging connectors 62, shown in
Vehicle 2 includes LEDs representing the head-, turn-, and break-lights, and special lights in unique vehicles such as police cars or firetrucks. Operation of those lights is similar to the operation of lights in a real vehicle. Microcontroller 40 controls these lights based on commands received from basestation 22 to show intended actions like turning left, braking, high beams, etc. The last part of the secondary input/output system can be an audio speaker which allows vehicle 2 to generate accoustic signals like honking, motor sounds, yelling drivers, sirens, etc. under the control of basestation 22 via wireless network radio (radio transceiver) 42 and microcontroller 40.
2.2 Vehicle Control Software:
Microcontroller 40 on each vehicle 2 operates under the control of vehicle control software to control the low level, real-time portion of the vehicle behavior, while the high level behaviors are controlled by basestation 22. The vehicle software operates in real-time with sub-system execution occurring within fixed periods or time slots. This means microcontroller 40 executes various tasks such as measuring speed, commanding motor voltage, steering, and checking for messages at different intervals. Hardware timers on microcontroller 40 are used ensure that each task is executed as desired. For example, each microcontroller 40 can take a scan using the imaging system at frequencies between 500 Hz-1000 Hz, command new motor speeds at frequencies between 250 Hz-500 Hz, check for new messages at 100 Hz and check the battery voltage every 10 seconds. Some tasks take longer than others to execute, but they all should execute within their allotted time periods or slots. The vehicle software can be divided into the following sub systems:
2.2.1 Communication:
Each vehicle 2, via its microcontroller 40 and wireless network radio 42, communicates wirelessly via messages with basestation 22 (which also includes a wireless transceiver) and reacts to messages sent by basestation 22. Messages sent by basestation 22 can, for example, but without limitation, include a new desired speed, a new desired acceleration, a lane switch command, request battery voltage, request the latest position information of the vehicle, turn on a turning signal, output a sound, etc. Vehicle 2 can also send messages about its own status without a request from basestation 22. This can happen when, for example, without limitation, the battery voltage is very low or vehicle 2 loses track of its position.
2.2.2 Marking Recognition/Interpretation:
The raw scans of markings 12 taken by imaging system 3 (where each pixel includes a grayscale value in the range of 0 to 255) are parsed by microcontroller 40 into bit values so that piece and location IDs can be computed. This happens through a series of steps shown in
Next, any one or more of a number of techniques can be used to classify the raw pixels of the scan into either black or white (
Once the scan is classified into white and black pixels, the resulting groups of black pixels can be inspected with error-checking techniques to correct for isolated errors and classified into the appropriated type of bar using expected widths of pixels in order to identify the center-line 10 and encoded values in a marking 12 row.
2.2.3 Steering Control Algorithm:
Each vehicle 2 needs to be able to steer precisely and at high speeds to follow a lane (e.g., without limitation, center-line 10) on a road piece 6. Microcontroller 40 on vehicle 2 uses imaging system 3 to not only identify markings from the road pieces 6, but also to compute their horizontal position within the field of view of the imaging sensor 46 of imaging system 3 at a high frequency. Unless otherwise instructed by basestation 22, microcontroller 40 is programmed to try to keep vehicle 2 centered over center-line 10 as seen in an immediately preceding scan by imaging system 3. Microcontroller 40 will compute the position of markings 12 relative to the center of vehicle 2, and if vehicle 2 is not centered, will cause vehicle 2 to steer as needed to move the markings 12 toward the center of vehicle 2. An Open/Closed Loop algorithm is used by microcontroller 40 to achieve that goal.
The Closed Loop portion of this algorithm is a PID (Proportional-Integral-Derivative) control which computes the steering angle for the current position error. The Open Loop portion of this algorithm uses prior knowledge embedded in markings 12 on road pieces 6 to determine whether the vehicle 2 is about to traverse a straight road piece (
If instructed by basestation 22, microcontroller 40 can choose to not center vehicle 2 on road markings 12, but execute a completely free trajectory. This is for example used when vehicle 2 switches from one road lane (e.g., center-line 10) to another.
This capability defines a difference between the present system and prior art toy slot cars or model railroad systems: namely, while prior art toy slot cars or model railroad systems are locked to a physical slot or track and cannot leave it, the present system's vehicles 2 can follow road markings 12 for some time, but can leave them at any time and make frequent use of that capability.
2.2.4 Speed Control Algorithm:
Microcontroller 40 is also responsible for driving vehicle 2 at a desired speed. Microcontroller 40 utilizes an open loop/closed loop speed control algorithm for speed control. The Open Loop speed control part commands an open loop speed to the motors 58 which will make the vehicle 2 drive at approximately the correct forward speed. The parsed road markings 12 acquired by the vehicle imaging system 3 are used to measure the amount of time it took vehicle 2 to travel over known lengths of the parsed road markings 12. This is converted by microcontroller 40 into the actual current forward speed of vehicle 2. Then, a closed loop (PID) speed control is used by the microcontroller 40 to eliminate the difference between the desired/commanded speed and the current speed of the vehicle 2.
As described above, forward speed estimation is performed by measuring the length of time vehicle imaging system 3 of vehicle 2 sees a row of markings 12. Since the length of each of these rows is known, the measured length of time for which the bars comprising each marking 12 are seen can be translated to an estimated speed of vehicle 2.
It is also desirable to measure lateral speed during lane changes to enable the speed of lane changes to be specified and controlled, allowing more accurate and diverse plans that involve both smooth and sharp lane change decisions through improved predictability of the vehicle's 2 motion. Also, this reduces the possibility of switching an incorrect number of lanes by always tracking the lateral distance traveled so far. This allows planning of large actions across many lanes at once rather than making a series of small single-lane transitions due to uncertainty concerns.
Rather than tracking the duration of a seen marking 12 as in the forward-motion case, lateral speed estimation works by tracking the lateral motion of the center-lines 10 of all markings 12 visible by the vehicle imaging system 3 throughout the lane transition. At every scan by the vehicle imaging system 3, the pixel positions of all visible bars' center-lines 10 are computed and compared against the positions from the previous scan. At a high enough scan frequency, these positions will move a maximum of a few pixels between scans, allowing microcontroller 40 to accurately match bars from consecutive scans. Microcontroller 40 tracks the overall progress of a reference center-line 10 and switches to another center-line 10 which becomes the new reference center-line 10 as soon as the current reference center-line 10 leaves the field of view, allowing uninterrupted lateral tracking throughout the entire motion.
By trading off between center-lines 10 as they pass through the field of a view of vehicle imaging system 3, microcontroller 40 can estimate an overall amount of horizontal translation in numbers of pixels from some starting point, even when the total translation is much greater than the width of the vehicle imaging system's 3 field of view. Since the distances between lanes are known, microcontroller 40 can execute a lane change at a given lateral speed by adjusting the heading of vehicle 2 by minimizing the difference between a calculated/determined lateral progress and a desired lateral progress at each point in time. If vehicle 2 is falling behind in its lateral progress, microcontroller 40 can cause vehicle 2 to turn sharper to catch up and if vehicle 2 is shifting laterally too quickly, microcontroller 40 can cause vehicle 2 to straighten its heading relative to the road piece 6 to slow down its lateral progress and reduce error.
This provides a high degree of accuracy in lateral speed control since at least one center-line 10 is visible in a majority of locations on the road pieces (note that fully parsing the marking row is not necessary for this computation). In a minority of situations where no markings 12 or center-lines 10 are visible, the current lateral motion can be estimated from the last-measured rate of lateral motion.
An example of this latter approach is shown in
Precise lateral motion execution through techniques such as this is necessary to enable vehicle 2 to execute maneuvers such as the one shown in
2.2.5 Motion Control Flow Algorithm:
A high level flow chart describing the control algorithm implemented by microcontroller 40 of each vehicle 2 is shown in
The control algorithm includes both steering and speed control and all the components necessary to gather the necessary information to correctly steer and move vehicle 2. Every cycle starts at step 82 by taking a scan using imaging system 3. In step 84, this scan from step 82 is analyzed and interpreted. If the scan is invalid, an error message may be sent to basestation 22 and vehicle 2 may use information from past scans to drive until a valid scan is recognized by microcontroller 40 or microcontroller 40 gets instructed otherwise by basestation 22. If the scan is valid, meaning it includes successfully-parsed road markings 12, in step 84, the next action for the vehicle is chosen based on this scan and the current state of vehicle 2.
In step 84, if vehicle 2 determines that the scan is invalid or if vehicle 2 cannot determine its next step, the algorithm advances to step 85 wherein execution of the algorithm is stopped and vehicle 2 executes a stop, shutdown, pause, etc. In contrast, if vehicle 2 is in a lane following mode, the algorithm advances to step 86 where microcontroller 40 computes the center of the center-line 10 in the scan and uses it to compute a new steering angle to center vehicle 2 over the road markings 12. Doing this consecutively for a number of road markings 12 causes the vehicle 2 to follow a path described by those road markings 12. If, in step 84, it is determined that vehicle 2 is in open loop mode, the algorithm advances to step 87 where microcontroller 40 executes an arbitrary trajectory commanded by basestation 22. Such trajectory can include lane changing, open loop turns, or anything else. As soon as the open loop maneuver has been executed, the algorithm will repeat steps 80-84 and enter into lane following mode by advancing to step 86. As shown in
If, in step 86, microcontroller 40 determines that imaging system 3 has not identified center-line 10, the algorithm advances to step 88 where microcontroller 40 transmits a warning to basestation 22 via wireless network radio 42. Basestation 22 responds to this warning by transmitting steering control information to microcontroller 40 which advances to step 90 and executes steering control of vehicle 2 utilizing this information from basestation 22. On the other hand, if in step 86, microcontroller 40 determines that center-line 10 has been found, the algorithm advances to step 92 where a new steering angle is computed. The algorithm then advances to step 90 where microcontroller 40 executes the new steering angle. Thus, in step 90, microcontroller 40 can act on steering control information from basestation 22, or the steering angle determined by microcontroller 40 in step 92.
From step 90, the algorithm advances to step 94 where a determination is made whether imaging system 3 has reached the end of a marking 12. If not, the algorithm returns to step 80 as shown. On the other hand, if the end of a marking 12 has been reached, the algorithm advances to step 96 where microcontroller 40 executes the speed control algorithm discussed above.
The algorithm in
Depending on the last marking(s) 12 seen, microcontroller 40 can make higher level decisions. For example, after detecting a series of markings 12 describing a unique location on a road piece 6 of the drivable surface, microcontroller 40 can send this location information back to the basestation 22 to allow it to track the position of vehicle 2. Another example is determining from the markings 12 whether a road piece 6 is straight, or turns left or right, allowing vehicle 2 to steer to account for the expected road curvature without specifically having to communicate with basestation 22.
2.2.6 Secondary Control Software:
The vehicle control software can also manage several secondary tasks. For example, it can monitor the battery voltage of vehicle 2 and decides whether it is too low and notify basestation 22 or shut down operation of vehicle 2 in extreme cases.
The vehicle control software also includes a light module that controls the vehicle's LED headlights, turn signals and brake lights. The brightness of all LEDs is controlled by PWM (Pulse Width Modulation). The light module is an example of software with multiple control levels. In most cases, basestation 22 will interact with the light module like a driver using the light control in a real car. Basestation 22 can choose to use turn signals when turning, choose to turn on/off lights or high beams, while functions like brake lights work automatically whenever vehicle 2 slows down quickly (braking). Basestation 22 can also or alternatively take direct control over single lights and determine their state, like brightness, blinking frequency etc. This is not a realistic behavior, but is useful to suggest special messages to the user, for example when batteries are very low, the vehicle software is rebooting, during software updates, etc.
The vehicle software also includes a sound module that causes a PWM signal to be output to a speaker. The sound module can modulate various frequencies on top of each other to generate sounds ranging from simple beeping to realistic voices.
The vehicle software can also include a state module that keeps track of the last state of vehicle 2 and remembers the state even if vehicle 2 is turned off. This allows each vehicle 2 to maintain data and parameters like maximum speed, sound (such as honking or sirens), its unique identifier (such as a license plate), etc. without requiring changes to the vehicle software.
A bootloader can allow for wireless software updates to each vehicle 2. Basestation 22 can initiate a software update and transmit the new software to a specific vehicle 2 or to all vehicles 2 simultaneously.
3 Non-Vehicle Agents:
Non-vehicle agents can also exist in the system. These can include, without limitation, stop lights 8, railroad track crossings, draw bridges, building garages, etc. Each of these non-vehicle agents can share the same general operating and communications structure as the vehicles: namely, each non-vehicle agent can have a microcontroller operating under the control of software to execute logic and behaviors, and act as another node in the system's network. This allows each non-vehicle agent to be represented and reasoned about within basestation 22 as with all other vehicles 2 and agents.
4 Basestation:
With reference to
4.1 Hardware:
Next the core components of basestation 22 will now be described.
4.1.1 Embedded Computer:
Basestation 22 includes an embedded computer (controller) that hosts the main software including, without limitation AI software, communications software, etc. Desirably, the computer is a small embedded system, for example an Intel Atom, ARMS, etc., with enough memory and clock speed to handle the algorithms within the software and to scale to a reasonably large number of vehicles 2 and other agents. The computer may also host a real-time/near real-time operating system like embedded Linux, XWorks, QNX, uLinux or similar. The foregoing description, however, is not to be construed as limiting the invention since it is envisioned that basestation 22 can be implemented by any suitable and/or desirable combination of hardware, operating system, and software now known in the art (e.g., without limitation, a game console) or hereinafter developed that is/are capable of implementing the functions of basestation 22 described herein.
Like each vehicle 2, basestation 22 hosts a wireless module/transceiver 100 (for example ZigBee, Bluetooth, WiFi, SimpliciTI, or similar) to allow communication with each vehicle and/or agent, e.g., via the wireless network radio 42 of vehicle 2. The only difference is that basestation 22 is the communication coordinator, while vehicles 2 are the end devices.
4.1.2 User Interface:
Basestation 22 may have a simple user interface 102 that includes buttons and switches (not shown) for user input and LEDs and, optionally, an LCD screen (not shown) for feedback to the user. User interface 102 enables the user to control high-level functions. For example, if vehicle-based exploration is being utilized to detect the drivable surface, user interface 102 enables the user to cause basestation 22 to initialize this exploration. Another example would be a general start-stop interface to initialize or terminate operation.
4.1.3 PC Connection:
It is possible to connect basestation 22 to a PC or laptop 106 via a PC connection 104, such as a USB. In this case, the user can have more control over functions of the system as well as improved user feedback, for example, via a RoadViz visualization application described herein. Via the software on PC 106, the user can adjust various parameters of the system, such as, without limitation maximum vehicle speed, vehicle behaviors, drivable surface behaviors, etc.
4.2 Software:
4.2.1 Visualization Tool:
Basestation 22 communicates with a RoadViz visualization application that can run on PC 106 using a network interface (e.g., protocol TCP/IP or UDP/IP). Basestation 22 sends messages to the RoadViz visualization application that update the system state, such as, without limitation, the structure of the drivable surface and the vehicle positions and plans. The communications protocol described later herein details some example messages that are passed between basestation 22 and the RoadViz visualization application running on PC 106.
4.2.2 Vehicles and Agents:
Basestation 22 communicates with vehicles 2 and other agents using a wireless network (such as Zigbee or Bluetooth) or a wired network in the case of static agents, e.g., traffic light 8, connected to road pieces 6. Desirably, the wireless module/transceiver 100 is connected to basestation 22 using a standard RS-232 serial interface. An attention (AT) command set is used to send/receive messages to/from specific vehicles 2 and agents.
When a new vehicle 2 or agent is introduced to the system, it must register with basestation 22 so it can be modeled within the system and controlled. When the new vehicle 2 or agent is started, it will send a message to basestation 22. Basestation 22 can also send a broadcast message to the entire network to query all present vehicles 2 and agents (for example, during initialization). Desirably, a special identification system is used so that multiple basestations 22 can be used in proximity to each other, and vehicles 2 must choose to register with a specific Basestation 22. In any event, each vehicle 2 is a unique node in the communications network that has a unique address that allows basestation 22 to uniquely communicate with the vehicle 2.
4.2.3 Messaging Library:
Basestation 22 includes a software module that facilitates communication with vehicles 2 and other agents via wireless transceiver 100 and manages message processing and delivery. This software module has several components. The first component, serialComms, is used to read and write data to/from a serial port of basestation 22. This module provides functions that abstract the specific transport layer of communications. The second component, carComms, is used by basestation 22 to formulate and send messages to vehicles 2 and other agents. The module will also keep a message mailbox for each vehicle 2 and agent, and will process incoming messages and deliver them to the appropriate mailbox. The third component, carMessages, is used to instantiate the specific messages. These components provide basic storage and serialization capability. The carComms module will instantiate a message using carMessages, and then send it using the serialComms component.
4.2.4 Artifical Intelligence Algorithm:
All interactions and high-level behaviors of the system are governed by the algorithms expected by basestation 22. This includes where vehicles 2 want to drive, how they plan to get there, how they interact with other vehicles 2 on the drivable surface, whether they follow traffic rules, etc. Basestation 22 can control both physical and simulated agents. The only difference is that the objects in the system representing physical agents (e.g., vehicles 2 and agents, such as traffic signal 8), send and receive real messages, while simulated agents interact with a software layer that simulates the responses and location updates from a physical agent. Both appear identical to basestation 22, allowing complex hybrid simulations with combinations of real and simulated agents.
Desirably, while each vehicle 2 executes all behaviors, it only directly controls low-level behaviors such as speed control, maintaining headings within a lane, and transitioning laterally to adjacent lanes. All higher-level planning described is computed entirely within basestation 22 and relayed to vehicle 2 which executes these plans through a series of simpler behaviors.
Much of the behavior in the system is driven by randomness (vehicle destinations, some behaviors, etc.). Desirably, basestation 22 is able to reproduce behavior in a fully simulated system of agents by using a deterministic random number generator that runs off of a seed value. This seed value can be initialized to produce random behavior (from the system clock for example) or to a previously used seed value to perfectly replicate the behavior of the system during that run in order to investigate any problems that arose.
4.2.4.1 Road Piece Network Representation:
Before initiating normal operation, basestation 22 must have a representation of the drivable surface that is being used. This can either be read in from a file accessible to basestation 22 or determined by basestation 22 at run-time using one of the methods described above.
With reference to
Along with a representation of the system state (such as drivable surface structure), each vehicle 2 and each static agent (such as traffic light 8) is represented in the virtual representation as an object that includes all information relevant to that agent. At specified frequencies, each vehicle 2 and/or static agent is presented by basestation 22 with the relevant information regarding the state of the rest of the system and told to update basestation 22 with its state, in effect making a decision concerning its behavior for the next time step (time period). This structure allows the processing for vehicles 2 and/or static agents to be parallelized across multiple threads, if desired.
4.2.4.2 Global Planning Algorithm:
A global planning algorithm is the core of basestation 22. All vehicle behavior is handled by modules that are called the global planner and the local planner. The global planner is responsible for high-level, long-term decisions such as determining the series of drivable road pieces 6 that need to be traversed in order to reach some destination in the drivable surface. For example, it might determine that the most efficient way for vehicle 2 to get from one point to another would be to take a U-turn followed by a left turn at the next intersection. The global planner abstracts away all local complexity such as lane changing, signaling, speed control and interactions with other vehicles and only considers the problem at the global scale.
While in many path planning applications a grid might be used to search for paths (where each square is connected to all its neighbors, representing possible motions), the present system has additional structure in the form of drivable sections (road pieces 6) that it can take advantage of to perform effective planning at a higher level. The global planner computes global paths by operating on the directed graph structure described earlier. Each edge in the graph includes a cost for traversing that drivable section. That cost is a function of various parameters such as length, maximum speed, number of lanes, and the presence of stop signs and lights, and could be customized for each such agent depending on their priorities. The global planner uses this weighted, directed graph to compute optimal paths using a graph search algorithm such as the A* or Dijkstra's algorithm.
The difference between A* and Dijkstra's algorithm is that A* uses a heuristic function that estimates the total cost from any state to the goal to guide the direction of the search. Since a reasonable estimate can be made for this cost from any state, A* is more desirable for this application. The A* algorithm traverses various paths from start to goal and for each node x, it maintains three values:
g(x): the smallest path cost found from the start node to node x;
h(x): the heuristic cost from node x to the goal; and
f(x)=g(x)+h(x)=the estimated cost of the cheapest solution through node x;
Starting with the initial node, A* maintains a priority queue of nodes to be explored, known as the open set, sorted in order of increasing value of f(x). At each step, the node with the lowest f(x) value is removed from the queue to be evaluated (since the goal is to find the cheapest solution), the g and f values of its neighbors are updated to reflect the new information found, and those neighbors are added to the open set if they had not been previously evaluated or if their f values have decreased from previous evaluation, representing a possibility of a better path through that node. In effect, the A* algorithm searches in the direction which appears to be best, often resulting in the optimal path with a much smaller amount of work compared to a brute-force search.
A heuristic is considered admissible if it is guaranteed not to overestimate the true cost to the goal. In such a two-dimensional path planning example, if the cost of a path were equal to the distance, the simplest admissible heuristic is the straight-line distance to the goal. With an admissible heuristic, once a path to the goal is found whose cost is lower than the best f(x) on the priority queue, it is guaranteed to be the optimal, lowest-cost path. Dijkstra's algorithm is equivalent to a special case of A* where h(x)=0 for all states.
For a simple example of A* search, see
With reference to
In
In
In
A* can easily be extended to search to a set of goal nodes rather than a single goal node by adjusting the heuristic function to estimate the optimistic cost to any goal node. Also, while the example of
4.2.4.3 Local Planning Algorithm:
Basestation 22 includes a local planning algorithm that executes the steps that enable vehicles 2 to execute a global plan. This includes speed control, distance keeping with other vehicles, lane changing decisions, behaviors at intersections, and signaling. For realism and scalability, decisions for vehicles 2 are made using only local knowledge rather than with full knowledge of the system to mirror real-world logic and allow the complexity of the system to scale with many vehicles 2 in a tractable way. For example, an object representing a vehicle 2 has full knowledge of its own state and plan but cannot use other vehicles' 2 plans in making its decision. It only has access to aspects of the system state that would be visible in the respective real-world scenario (locations and speeds of nearby vehicles 2, the state of traffic lights 8, etc.).
4.2.4.3.1 Intersection Behavior:
With reference to
4.2.4.3.2 Speed Control:
The speed-related high-level computations by basestation 22 desirably assume a fixed acceleration and, therefore, use the simple model illustrated in
Here a vehicle changes from an initial velocity Vi to a final velocity Vf over time t with an acceleration of a. The distance, d, over which this speed change will take place can be determined by integrating the area under this function as follows:
There are numerous scenarios when a variable needs to be computed and other variables are known. For example, if it is desired to stop from an initial velocity Vi over time t, an acceleration of
is required.
With reference to
When this is executed at a relatively high frequency for each vehicle 2, basestation 22 is able to achieve smooth and realistic distance keeping in complex traffic systems through this computationally efficient and decentralized approach.
The same computation can be used to achieve a speed change such as stopping at a specific location: the vehicle 2 can compute the acceleration a that allows it to transition from its original speed Vi to a final speed Vf=0 over a remaining distance. However, due to communication delays, uncertainty of positioning and unpredictable speed changes due to traffic and other conditions, speed change commands within basestation 22 and vehicles 2 are specified relative to absolute locations identified by position markings 12 rather than commanded for immediate execution. For example, to stop at a stop sign or the end of a path, a vehicle 2 would be sent a command by basestation 22 to achieve a speed of 0 with a certain deceleration at an offset from some location markings 12 encountered earlier. Having an absolute location to track its position from, as vehicle 2 approaches the desired stopping point vehicle 2 will continually recompute travelDist, the distance required to achieve the target velocity at the specified acceleration, and will begin executing the speed change when travelDist is equal to the remaining distance to the destination. In this way vehicle 2 is able to execute a realistic stop at stop signs regardless of the traffic conditions in which it is driving.
4.2.4.3.3 Full Local Planning with Lane Changing:
The full local planning algorithm implemented by basestation 22, that includes speed and lane changing decisions, can be treated as a multi-dimensional planning problem rather than planning in a two-dimensional, position-based search space, as in the case of the global planning algorithm. One desirable approach used by basestation 22 is to treat this problem as a planning problem in four dimensions:
These dimensions form the search space where a point in this space corresponds to a state, i.e., some value for each of the dimensions mentioned above. Each point in this space connects to other points in this space representing states that are reachable from the state after some action. For example, a point in this space may have a connection to another point representing adjacent lanes forward in distance and time relative to its speed, but not to points representing lanes far away or times in the past (since these transitions are not possible). So in effect, this forms a graph search problem as with the global planning algorithm, except that the search space and branching factor are significantly larger. In fact, while there are a large number of possible states based on these four dimensions, only a relatively small subset of them are relevant for the search problem. The planning horizon, or how far into the future distance basestation 22 computes plans, is defined by the maximum value for the forward distance dimension under consideration. This is a trade-off between computational complexity (since a larger forward distance increases the size of the graph to be searched) and plan effectiveness (the need to plan sufficiently far into the future to generate intelligent plans).
One assumption to make for short planning windows concerning the paths of other vehicles 2 is that these other vehicles 2 will maintain their current speed and lane unless they are signaling otherwise. It is also possible to incorporate uncertainty about the motions of other vehicles 2 by penalizing states that have some potential to be affected by those vehicles 2. While the local plan is computed by basestation 22 far enough into the future to identify sophisticated behaviors, the local plan will be recomputed frequently, allowing basestation 22 to react to any deviations from the assumed behavior of vehicles 2.
Basestation 22 solves this multi-dimensional planning problem by planning from the starting point in this space to any point at the planning horizon (all nodes with the specified forward distance dimension value are considered goal states) subject to some optimization function. Such a function could, for example, penalize lane or speed changes and closer encounters with moving obstacles, and reward higher speeds, progress towards a goal or being positioned in a specific lane if a turn is planned in the future. The function being optimized captures the current goal of the vehicle 2, and the goal of basestation 22 is to find a series of allowable actions through this high dimensional space that optimizes the value accrued from this function.
As mentioned previously, while this multi-dimensional state space can be large and difficult to fully represent in a memory of basestation 22, the entire space does not need to be represented since most states will never be considered. One desirable, non-limiting, implementation can allocate space for new states only as they are considered, reducing the memory requirement to only the relevant fraction of the full state space.
Basestation 22 can use optimal algorithms, such as A* or Dijkstra's, or can utilize sampling based probabilistic approaches such as Rapidly-Exploring Random Trees (RRTs) biased towards the planning horizon since plans do not need to be optimal and the search space may be too large. In such approaches, the aspects of the state space no longer need to be discretized at a specific resolution since sampling techniques can operate on arbitrary locations in the state space.
Another option is to utilize a special version of A* called Anytime Repairing A* (ARA*).ARA* has the property that it will first quickly find a sub-optimal solution to the planning problem and will spend the remaining time iteratively improving it as time permits. ARA* accomplishes this by repeatedly calling the normal A* algorithm but multiplying the values returned by the heuristic function by some constant ε>1. This new heuristic is no longer admissible (since it may overestimate the true cost to the goal from any state), but it reduces the search time significantly since fewer states will appear to have a possibility of contributing to the path. Even though the final solution will no longer be optimal, it is guaranteed to have a cost at most ε times larger than the true optimal cost and in practice is often much closer to the optimal. By gradually reducing and intelligently reusing much of previous iterations' computations, ARA* has the desirable property that a reasonable solution, often close to the optimal, will always be available within the fixed time that the algorithm has to operate. This allows basestation 22 to compute high-quality plans while guaranteeing that its required update frequencies will be met.
To better understand the purpose of this local planning by basestation 22 and the types of plans that may be found by such an approach, consider the example plan shown in
4.2.4.3.4 Other Logic:
Additional logic within basestation 22 controls behaviors such as logic for traffic light 8 signaling and execution of sounds. Intended behavior is communicated by basestation 22 via messages to the physical agents for execution.
4.2.5 Basestation Software Summary:
A flow diagram explaining possible logic in basestation 22 at a high level is shown in
5 User Interface:
Vehicles 2 can be controlled by basestation 22 or by a user, for example, via a remote control 132. A user's main interaction tools with the system is a remote control 132, a PC 106 connected to basestation 22, or both a remote control 132 and a PC 106.
5.1 Remote Control:
Each remote control 132 includes a wireless transceiver (not shown) which is part of the system's wireless network. As with vehicles 2, there can be many remote controls 132 interacting with the basestation 22 and vehicles 2 simultaneously. In the most common case, each remote control 132 is used by a user to control a specific vehicle 2. What vehicle 2 is being controlled can be changed at any time by the user. When the user switches control away from one vehicle 2, basestation 22 resumes full control over that vehicle 2. All vehicles 2 not controlled by a remote control 132 are automatically controlled by basestation 22. Compared to common remote controlled toy cars, the remote controls 132 described herein offer a higher level of interaction. The steering itself is desirably performed by the vehicle 2 and not by the user, since vehicles 2 can move quickly and the roads can be narrow. A user can instead provide higher level commands to a vehicle 2 in the form of speed adjustments, deciding to switch lanes, deciding where to turn at intersections, initiate U-turns, pass other vehicles 2, etc. Also, a user can have control over secondary vehicle functions like turning signals, lights, honking, etc.
In addition to controlling vehicles, remote controls 132 can also be used to control stationary agents, such as the special components: traps on road pieces, traffic lights 8, road barriers, garage doors, etc.
5.2 PC Control:
The controls described above in connection with remote control 132 can also or alternatively be replicated through PC 106 using a keyboard, a mouse and/or an attached gamepad. Additionally, this allows the possibility of an operator commanding a vehicle over the internet using a visualization of the system state.
6 Visualization Software:
With reference to
6.1 Components:
6.1.1 Graphics Engine:
Desirably, the visualization application uses a 3D graphics engine. One implementation can be built using C# and the Microsoft XNA platform. Via the visualization application, the user can control a virtual camera to view the network at a desired location. An example screenshot on visual display 140 is shown in
6.1.2 Software Structure:
The visualization application desirably uses several threads to distribute its computational load. One application thread is dedicated to rendering the graphics, and another thread is for network communications with basestation 22.
6.1.3 Communications:
The visualization application desirably uses a network socket communications (such as TCP/IP) and acts as a server. Basestation 22 (or similar) agent can connect to the visualization application running on PC 106 (
Basestation 22 can then send and receive messages via the visualization application. Some exemplary messages are discussed hereinafter.
6.2 Debugging Abilities:
The visualization application is useful for debugging basestation 22. The visualization application receives system state information and can request or send information back to basestation 22. The visualization application desirably includes a menu system from which a user can view the drivable surface or send/receive this specific information.
7 Communications Protocol:
Next, the communication protocol between various components of the system will be described with reference to a sampling of the various messages that can be used the system.
7.1 Basestation—Visualization Application Tool:
7.1.1 Message Descriptions:
CLEAR_MAP—resets the state of the visualization application and removes all road pieces, vehicles, etc.;
DISPLAY_ROAD_PIECE—commands the visualization application to display a particular type of road piece at a particular location;
CREATE_CAR—commands the visualization application to display a particular type of vehicle at a particular location;
SET_CAR_POSE—commands the visualization application to update the position and orientation of a vehicle;
DISPLAY_PATH_PLAN—commands the visualization application to display the planned path of a vehicle;
SET_STATE—changes the state of a variable in the basestation as requested by the visualization application user;
REQUEST_STATE—requests information from the basestation to be displayed by the visualization application on visual display 140.
7.2 Basestation—Agents (Vehicles, Etc.):
7.2.1 Message Descriptions:
CMD_LIGHTS—basestation 22 instructs a vehicle to set its lights on, off, blinking;
CMD_LINARRAY_DATA_REQUEST—basestation 22 instructs a vehicle 2 to send data from a linear array scan by imaging system 3;
CMD_LINARRAY_DATA_RESPONSE—vehicle 2 sends the linear array scan data to basestation 22;
CMD_SCANLED—basestation 22 instructs vehicle 2 to turn its scan LED 44 on/off;
CMD_SET_EXPOSURE—basestation 22 instructs vehicle 2 to set the exposure time of the linear array of imaging system 3;
CMD_BATTERY_VOLTAGE_REQUEST—basestation 22 requests a vehicle's battery voltage;
CMD_BATTERY_VOLTAGE_RESPONSE—vehicle 22 sends its battery voltage to basestation 22;
CMD_SHIFT_LANE—basestation 22 instructs a vehicle 2 to shift to another lane;
CMD_SET_SPEED—basestation 22 instructs a vehicle 2 to achieve a certain speed;
CMD_PING_REQUEST—basestation 22 sends a vehicle 2 (or all vehicles) to respond with an “alive” (ping) notice;
CMD_PING_RESPONSE—vehicle 2 sends an “alive” (ping) response to the basestation 22;
CMD_POSE_REQUEST—basestation 22 requests the pose information from the vehicle 2 at a particular frequency;
CMD_POSE_UPDATE—vehicle 2 sends its pose information to the basestation;
CMD_BRANCH—basestation tells vehicle to follow a branch through an intersection (left, right, straight).
7.2.2 Typical Usage:
Next, a full sequence of messages that are passed between a vehicle 2 and basestation 22 during a complex driving maneuver will be described. The maneuver involves a vehicle 2 reporting its pose (where “pose” means a vehicle's position x, y and heading theta) during normal driving, changing lanes from left to right, stopping at an intersection, and then making a right turn and resuming normal driving on a new drivable section. It is important to notice that vehicle 2 doesn't actually have to send the full pose information to the basestation, but just parsed scans of marking 12 by vehicle imaging system 3, which basestation 22 uses to derive vehicle's 2 pose. This greatly reduces the need for computational power on the vehicle 2.
7.2.3 Dealing with Uncertainty:
Although the wireless transceivers of the system will guarantee the delivery of messages, there is some uncertainly as to the delivery time of these messages. The messaging protocol does not guarantee message delivery within any specified amount of time and, as a result, there is some amount of uncertainty of the lag between when a message is sent and when it is received. Thus, both basestation 22 and the vehicles 2 must account for this uncertainty.
Fortunately, vehicle 2 is responsible for the low-level control (lane following, etc. . . . ) and basestation 22 only needs to send high-level controls to vehicle 2. This allows basestation 22 to send messages well before they need to be acted upon by vehicle 2. For example, a speed change command will instruct vehicle 2 to change speed after reaching a certain location. Vehicle 2 receives this message potentially several seconds in advance, and then takes the appropriate action when it needs to (e.g., change speed after a certain marking 12 is read by vehicle imaging system 3).
Further, basestation 22 can create path plans for vehicles 2 that account for uncertain timing in the message delivery. Vehicles 2 will maintain a safe distance behind other vehicles 2 so that they will have ample time to receive messages and act upon them.
Basestation 22 will also forward simulate the vehicle's 2 position. Since basestation 22 knows the speed of vehicle 2, it can estimate the vehicle's 2 actual position between receiving pose updates through the CMD_POSE_UPDATE message. This knowledge, along with statistics of message delivery times can be used to better estimate when messages should be sent so they can be received and acted upon in a timely manner.
8.0 Marker Obfuscation Through IR Transparent Film:
As described above, a series of markings 12 enables vehicles 2 to identify their unique positions in the drivable surface. A technique described above for encoding markings in a way not visible to humans relied upon printing the markings in an ink or dye that is not visible (transparent) to the human eye and absorbs light in the IR, NIR, or UV spectrum. By using a light source of the same light wavelength, these markings appear black to the optical sensor but are nearly or completely invisible to humans.
Alternatively, markings 12 can be printed in standard visible ink or dye, for example used in commercial inkjet printers, laser printers or professional offset or silk screen printing machines. After the markings are printed, a second layer is applied to cover those markings. This second layer includes an ink or dye or a thin plastic film that is transparent above or below human visible wavelengths, but appears opaque in the human visible spectrum. Materials having such properties are commercially available.
For the purposes of this example, near infrared (NIR) light with wavelength of approximately 790 nm will be discussed, but the same approach can be used for any non-visible portion of the light spectrum (UV, IR, NIR, etc.). When this surface is used with a vehicle imaging system 3 with an NIR responsive imaging sensor 46 under NIR light from LED light source 44, the light will pass through the NIR-transparent material, allowing the optical sensor to detect the markings 12 underneath (most standard black ink/dye will still appear black to the optical sensor under NIR light). To the human eye, only the surface material will be visible since light in the visible spectrum will not be able to pass through. An illustration of this approach as well as the appearance to the human eye of a segment that uses this approach can be seen in
Such an approach provides an advantage in terms of ease of manufacturing and potentially low cost because codes can then be printed using standard ink or dye and standard printing techniques (ink jet, laser, offset printing machines, silk screen printing machines, etc.). Material transparent to non-visible light can then be applied using any method. Some examples include: printing, film, spray paint, stickers, decals, etc. This can also allow users to print their own drivable surfaces and then simply apply the transparent material to the surface using any of the techniques mentioned.
It is envisioned that a software application (through a PC or web-based interface) can be provided that enables a user to design a drivable surface according to their personal specifications. For example, the user can develop large-format (e.g. 12 ft×30 ft) drivable surfaces that use custom designed drivable segments. Users can specify any road piece shape they desire that includes combinations of straight segments and arcs of circles (each segment could be required to be of some minimum length) or even more complex shapes like splines etc. The software application then processes the final network shape and decomposes it into the necessary combination of segments.
This custom drivable surface can then be manufactured for the user using a flexible material, such as vinyl, that can be rolled up, transported and stored easily, taking up only a fraction of the space necessary compared to a similar drivable surface made out of rigid plastic parts. The drivable surface will appear visually the same as how the user designed it, but will also contain the position identification markings which are hidden from view using the techniques described above.
In addition to the final drivable surface, the user can also be provided with a file defining that particular network. The file can be transferred to the user's basestation 22 so that the basestation 22 can interact with the unique structure of that drivable surface by identifying the unique positions that each set of markings encodes and allowing vehicles 2 to generate plans accordingly.
The invention has been described with reference to exemplary embodiments. Obvious modifications and alterations will occur to others upon reading and understanding the preceding detailed description. It is intended that the invention be construed as including all such modifications and alterations insofar as they come within the scope of the appended claims or the equivalents thereof.
This application claims priority as a continuation of U.S. Utility application Ser. No. 14/574,135, for “Distributed System of Autonomously Controlled Mobile Agents” (Atty. Docket No. ANK001CONT5), filed on Dec. 17, 2014, which is incorporated herein by reference. U.S. Utility application Ser. No. 14/574,135 claimed priority as a continuation of U.S. Utility application Ser. No. 14/265,092, for “Distributed System of Autonomously Controlled Mobile Agents” (Atty. Docket No. ANK001CONT3), filed on Apr. 29, 2014, and from U.S. Utility application Ser. No. 14/265,093, for “Distributed System of Autonomously Controlled Mobile Agents” (Atty. Docket No. ANK001CONT4), filed on Apr. 29, 2014, both of which are incorporated herein by reference. Both U.S. Utility application Ser. No. 14/265,092 and U.S. Utility application Ser. No. 14/265,093 claimed priority as continuations of Utility application Ser. No. 13/707,512, for “Distributed System of Autonomously Controlled Toy Vehicles” (Atty. Docket No. ANK001CONT), filed on Dec. 6, 2012 and issued as U.S. Pat. No. 8,747,182 on Jun. 10, 2014, which is incorporated herein by reference. U.S. Utility application Ser. No. 13/707,512 claimed priority as a continuation of U.S. Utility application Ser. No. 12/788,605, for “Distributed System of Autonomously Controlled Toy Vehicles” (Atty. Docket No. ANK001), filed on May 27, 2010 and issued as U.S. Pat. No. 8,353,737 on Jan. 15, 2013, which is incorporated herein by reference. U.S. Utility application Ser. No. 12/788,605 claimed priority from U.S. Provisional Patent Application Nos. 61/181,719, filed on May 28, 2009, and 61/261,023, filed on Nov. 13, 2009, both of which are incorporated herein by reference.
| Number | Date | Country | |
|---|---|---|---|
| 61181719 | May 2009 | US | |
| 61261023 | Nov 2009 | US |
| Number | Date | Country | |
|---|---|---|---|
| Parent | 14574135 | Dec 2014 | US |
| Child | 14964438 | US | |
| Parent | 14265092 | Apr 2014 | US |
| Child | 14574135 | US | |
| Parent | 14265093 | Apr 2014 | US |
| Child | 14265092 | US | |
| Parent | 13707512 | Dec 2012 | US |
| Child | 14265092 | US | |
| Parent | 13707512 | Dec 2012 | US |
| Child | 14265093 | US | |
| Parent | 12788605 | May 2010 | US |
| Child | 13707512 | US |
