This application claims priority to European Patent Application No. 19184425.7, filed Jul. 4, 2019, which is incorporated herein by reference.
The invention relates to methods to select one trajectory, the so-called Selected Trajectory, out of a set of trajectories to be used by an autonomous or semi-autonomous ground vehicle.
Furthermore the invention relates to a Selected-Trajectory selecting-device for selecting one trajectory and to a Selected-Trajectory generating-device.
Finally, the invention relates to an autonomous or a semi-autonomous ground vehicle comprising a Selected-Trajectory selecting-device for selecting one trajectory and/or a Selected-Trajectory generating-device.
The invention relates to autonomously or semi-autonomously ground vehicles (abbreviated by the term vehicle), such as autonomous or semi-autonomous automobiles.
Said vehicles operate in close proximity with other autonomous or non-autonomous vehicles in public infrastructures, such as highways, urban, or sub-urban settings. Said vehicles implement Trajectory Generators that generate trajectories along which trajectories the vehicle shall move. However, Trajectory Generators may become faulty and produce unsafe trajectories.
Unsafe trajectories are trajectories that are unsafe in a sense that the ground vehicle, if it was to move along said faulty trajectory, would cause harm to passengers within said vehicle or human beings outside said vehicle. For example, a trajectory that causes a crash of said vehicle with a stationary or moving object is considered an unsafe trajectory if it is reasonable to assume that the passengers inside the vehicle or human beings in the environment of said vehicle would become injured as a result of the crash.
Trajectories that are not unsafe are considered to be safe trajectories.
It is an objective of the invention to disclose a method and a device for Safe Trajectory Selection, that selects a trajectory, the so-called “Selected Trajectory” out of a set of trajectories generated by two or more Trajectory Generators, which can be used by a vehicle.
This object is achieved a method as mentioned above, wherein according to the invention said method is characterized by the following steps:
According to the method the best ranked trajectory may be or selected as Selected Trajectory.
Under the assumption that a least one Trajectory Generator produces safe trajectories, said Selected Trajectory is a safe trajectory, even in situations when one, two, or more of the Trajectory Generators produce unsafe trajectories.
In the case that two or more Trajectory Generators produce safe trajectories, according to the invention, the best safe trajectory is chosen as Selected Trajectory.
The Safe Trajectory Selection is configured to apply a Trajectory Ranking Scheme to determine the best safe trajectory as the Selected Trajectory. The particularities of said Safe Trajectory Selection as disclosed in this application significantly improve the probability that the autonomous or semi-autonomous vehicle manoeuvres along a safe trajectory which is the best trajectory beyond the current state-of-the-art.
Preferably, two or more Trajectory Generators are provided, which usually are implemented in a vehicle. All Trajectory Generators are configured to produce safe trajectories. However, due to failures (e.g., design failures, software failures, random hardware failures, etc.) one, two, or many Trajectory Generators may fail and as a result of this failure produce unsafe trajectories. The Safety Trajectory Selection according to the invention implements Verification Modules as well as a Decision Maker in order to select a Selected Trajectory from the trajectories generated by the Trajectory Generators, wherein the Selected Trajectory is the best trajectory with respect to a configured Ranking Scheme and preferably is safe.
As mentioned above, it may be provided that the set of trajectories is generated by two or more Trajectory Generators, wherein preferably each Trajectory Generator of the two or more Trajectory Generators generates one or more of the trajectories of the set of trajectories independently from the other Trajectory Generators.
One, two or more of the trajectory generators may fail and produce unsafe trajectories. However, using two or preferably more trajectory generators enhances the probability that at least one trajectory generator produces safe trajectories, so that the best trajectory with respect to the Ranking scheme is a safe trajectory.
It may be provided that all Trajectory Generators are diverse, in particular in that each of the Trajectory Generators uses different algorithms for generating trajectories then the other Trajectory Generators and/or each Trajectory Generator is implemented on different hardware.
In this case it can be possible to identify Trajectory Generators and/or algorithms which produce, in particular in specific situations, “better” trajectories and Trajectory Generators which produce “less good” or even unsafe trajectories, and to use this information about the Trajectory Generators and/or algorithms as additional input (see also below) for ranking the trajectories according to the Ranking Scheme.
The Verification Modules analyse the trajectories generated by the Trajectory Generators with respect to various properties. Examples of Verification Modules (the checks which said modules may execute) are: analysis whether the probability of collision with an obstacle is sufficiently low, an analysis whether the trajectory is drivable by the vehicle according to the vehicle dynamics, or an analysis whether the trajectory is in line with legal regulations.
The Decision Maker takes the results of the Verification Modules as input and preferably takes additional inputs IN as well. Additional input IN is for example any one or an arbitrary combination of the following: the Vehicle State information (velocity and/or acceleration and/or direction and/or tire friction and/or steering angle), Map Data, Trajectory Generator Diagnostics, knowledge about the Automotive Safety Integrity Levels (ASIL) of the Trajectory Generators, knowledge about the Vehicle Environment. Based on said input the Decision Maker selects the Selected Trajectory based on its input and using a Ranking Scheme.
The vehicle uses the Selected Trajectory as selected by the Safe Trajectory Selection to determine its future movement.
Preferably, the method and the vehicle is configured to execute the procedure as outline above (trajectory generating, trajectory selection, trajectory execution) periodically. Typical periods of this process execution are from 1 ms to 100 ms, but may also be well below 1 ms or well above 100 ms.
Further advantages of the invention, which alone or in any arbitrary combination may be realised, are described in the following:
Preferably, different verification modules execute different tests on the trajectories.
A Selected-Trajectory selecting-device for selecting one trajectory, the so-called “Selected Trajectory” out of a set of trajectories, which Selected Trajectory is to be used by an autonomous or semi-autonomous ground vehicle, according to the invention comprises one, two, or a multitude of verification modules and a Decision Maker, wherein the Safety-Trajectory-selecting device is adapted to execute a method as described above.
A Selected-Trajectory generating-device according to the invention comprises
Preferably, each of the trajectory generators is adapted to generate one or more of the trajectories of the set of trajectories independently from the other Trajectory Generators. It may be provided that all Trajectory Generators are diverse, in particular in that each of the Trajectory Generators uses different algorithms for generating trajectories then the other Trajectory Generators and/or each Trajectory Generator is implemented on different hardware then the other Trajectory Generators.
In the following, in order to further demonstrate the present invention, illustrative and non-restricting embodiments are discussed, as shown in the drawings, which show:
We discuss some of the many implementations of the invention next. If not stated otherwise, all details described in connection with a specific example are not only valid in connection with this example, but apply to the general scope of protection of the invention.
Furthermore, the Safe Trajectory Selection STS may have locally stored Information INF as well, as for example information about the Automotive Safety Integrity Level (ASIL) of the respective Trajectory Generators TG1-TG3, and/or diagnostics data on the operational history of the Trajectory Generators TG1-TG3, and/or other information, which may be used for selecting the Selected Trajectory ST.
The Safe Trajectory Selection STS further implements one or more Verification Modules, in this example four Verification Modules VM1-VM4. The Verification Modules VM1-VM4 are configured to implement tests on the trajectories T1-T3. Said tests on said trajectories T1-T3 return Quality Assessments Q11-Q43, wherein said Quality Assessments Q11-Q43 are indicating the quality of each of the trajectories T1-T3 in terms of said tests.
Preferably, each Verification Module implements a different test on one, two or preferably all trajectories.
For example,
The Safe Trajectory Selection STS selects a trajectory as Selected Trajectory ST from the trajectories T1-T3 generated by the Trajectory Generators TG1-TG3. In general, each Trajectory Generator generates one or more trajectories independently of the other Trajectory Generators. In the specific example, Trajectory generator TG1 generates trajectory T1, Trajectory generator TG2 generates trajectory T2, and Trajectory generator TG3 generates trajectory T3.
The selection is executed in a Decision Maker DM. The Decision Maker DM uses the Quality Assessments as returned by the tests in the Verification Modules VM1-VM4, and optionally other input IN (as described above), as well as optionally locally stored information INF for the purpose of selecting of the one Selected Trajectory ST. For this selection process, the Decision Maker DM encodes a Ranking Scheme RS according which the Decision Maker DM ranks the trajectories T1-T3. Thus, this Ranking Scheme RS uniquely identifies the best safe trajectory out of the set of trajectories T1-T3.
The Safe Trajectory Selection STS returns said best trajectory as output as Selected Trajectory ST. Said Selected Trajectory ST is typically a safe trajectory or the best safe trajectory under the assumptions mentioned in the beginning.
Example tests of Verification Modules VM1-VM4 are as follows:
In the following we give some example embodiments of Ranking Schemes RS. The Verification Modules VM1-VM4 are preferably of different criticality. For example the Verification Module implementing a Collision Probability Test CPT is more critical than the Verification Module that is configured to implement a Comfort and Convenience Test CCT, because a low Quality Assessment Q11-Q43 of a trajectory by the Collision Probability Test CPT means a high probability of collision of the vehicle GV with an obstacle, while a low Quality Assessment of a trajectory by the Comfort and Convenience Test CCT does not compromise the safety of vehicle passengers. Thus, in one embodiment the Ranking Scheme RS distinguishes between Safety-Relevant Verification Modules S-VM and Non-Safety-Relevant Verification Modules N-VM.
In one embodiment the Verification Modules VM1-VM4 are statically defined as Safety-Relevant Verification Modules S-VM and Non-Safety-Relevant Verification Modules N-VM for the duration of a complete mission of a vehicle GV.
In one embodiment the distribution of the Verification Modules VM1-VM4 into Safety-Relevant Verification Modules S-VM and Non-Safety-Relevant Verification Modules N-VM may change during the mission of a vehicle GV. It may be provided that said distribution changes during the mission of a vehicle GV depending on inputs IN that the Safe Trajectory Selection STS receives. Based on said inputs IN, the Safe Trajectory Selection STS may differentiate between driving scenarios, e.g., highway driving vs. parking vs. urban or suburban driving. Thus, the Safe Trajectory Selection STS may change the classification of the Verification Modules VM1-VM4 into Safety-Relevant Verification Modules S-VM and Non-Safety-Relevant Verification Modules N-V in response to changing driving scenarios.
It may be provided that the Safe Trajectory Selection STS normalizes the Quality Assessments Q11-Q43 to ranges, e.g. to a values RACT in a range between a minimum value RMIN and a maximum value RMAX.
It may be provided that the Safe Trajectory Selection STS normalizes two or more, preferably all Quality Assessments Q11-Q43 to the same range.
It may be provided that the Safe Trajectory Selection STS is configured to implement configurable weights WACT, within an interval of a minimum weights WMIN and maximum weights WMAX, for Quality Assessments Q11-Q43. It may be provided that the interval of configurable weights [WMIN, WMAX] is identical for two or more, in particular for all Quality Assessments Q11-Q43.
It may be provided that the configured weights WACT are static for the duration of a complete mission of a vehicle GV.
It may be provided that the configured weights WACT change during the mission of a vehicle GV. The configured weights WACT may change during the mission of a vehicle GV depending of the inputs IN that the Safe Trajectory Selection STS receives. Based on the inputs IN, the Safe Trajectory Selection STS may differentiate between driving scenarios, e.g., highway driving vs. parking vs. urban or suburban driving. Thus, the Safe Trajectory Selection STS may change the configuration of the weights WACT in response to changing driving scenarios.
The Ranking Scheme RS defines Ranking Procedures on how to calculate a Trajectory Rank TR1-TR3 for a respective Generated Trajectory T1-T3. The Safe Trajectory Selection STS selects the Trajectory T1-T3 with the best Trajectory Rank TR1-TR3 as the Selected Trajectory ST. In the following we discuss some examples of Ranking Procedures.
In one embodiment the Ranking Procedure RP is realized as follows:
SRi=Sum(WACT*Qji),
NRi=Sum(WACT*Qki),
This Tie Braker TB can be for example:
Number | Date | Country | Kind |
---|---|---|---|
19184425.7 | Jul 2019 | EP | regional |