Dynamics-Dependent Behavioral Planning for at least Partially Self-Driving Vehicles

Abstract

A method for training a behavior planner for an at least partially self-driving target vehicle on the basis of observation data regarding kinematics and/or dynamics that have been recorded during at least one test drive in a test vehicle includes identifying a driving maneuver that moves the test vehicle from an initial state to an end state using the observation data, ascertaining the maneuver end time, retrieving a maneuver duration required by the target vehicle to perform the identified driving maneuver from a dynamics model of the target vehicle, labeling observation data from a time interval, defined by the maneuver duration, with the identified driving maneuver, and training the behavior planner, using the labeled observation data, to map observation data that indicate a state of the target vehicle to at least one driving maneuver to be performed.

Claims

1. A method for training a behavior planner for an at least partially self-driving target vehicle on the basis of observation data relating to the kinematics and/or dynamics that have been recorded during at least one test drive in a test vehicle, comprising: identifying at least one driving maneuver that moves the at least one test vehicle from an initial state to an end state from the observation data; determining a maneuver end time at which the at least one test vehicle reaches the end state; retrieving a maneuver duration required by the at least partially self-driving target vehicle to carry out the identified driving maneuver from a dynamics model of the at least partially self-driving target vehicle; labeling observation data from a time interval, an end of which is defined by the maneuver end time and the start of which is before the maneuver end time by the maneuver duration, with the identified driving maneuver; and training the behavior planner using the labeled observation data, to map observation data indicating a state of the at least partially self-driving target vehicle to at least one driving maneuver to be carried out.

2. The method according to claim 1, wherein: the driving maneuver is identified with a trained classifier model; and said classifier model maps a time series of observation data to at least one driving maneuver consistent with said time series.

3. The method according to claim 1, wherein a dynamics model of the at least partially self-driving target vehicle is selected, which is configured to map a combination of an initial state of the at least partially self-driving target vehicle and a specification of a driving maneuver to a maneuver duration.

4. The method according to claim 3, wherein the combination additionally includes at least one adjustable parameter of the behavior planner of the at least partially self-driving target vehicle.

5. The method according to claim 1, wherein the at least one driving maneuver includes at least one of: a lane change to an adjacent travel lane; a change in a driving speed by a specified amount; stopping the at least partially self-driving target vehicle at a defined location; driving behind another vehicle; and emergency braking of the at least partially self-driving target vehicle.

6. A method, comprising: training a behavior planner for an at least partially self-driving vehicle by; identifying at least one driving maneuver that moves the test vehicle from an initial state to an end state from the observation data, determining a maneuver end time at which the test vehicle reaches the end state, retrieving a maneuver duration required by the target vehicle to carry out the identified driving maneuver from a dynamics model of the target vehicle, and labeling observation data from a time interval, an end of which is defined by the maneuver end time and the start of which is before the maneuver end time by the maneuver duration, with the identified driving maneuver; acquiring observation data relating to the kinematics and/or dynamics of the vehicle using at least one sensor of the at least partially self-driving vehicle; mapping the observation data, using the behavior planner, to at least one driving maneuver to be carried out; and controlling at least one actuator acting on the driving dynamics of the vehicle such that the vehicle carries out the driving maneuver.

7. A method for training a dynamics model of a target vehicle for use in the method according to claim 1, comprising: providing learning initial states and learning driving maneuvers; carrying out the learning driving maneuvers based on the learning initial states on the target vehicle and/or on a simulation model of the target vehicle; determining and defining the respective time required to carry out the learning driving maneuvers as the learning maneuver duration; feeding the learning initial states and learning driving maneuvers to the dynamics model; mapping with the dynamics model, the learning initial states and learning driving maneuvers to a respective maneuver duration; evaluating agreement of the maneuver duration with the learning maneuver duration associated with the respective learning initial state and learning driving maneuver using a specified cost function; and optimizing parameters characterizing the behavior of the dynamics model so that the further processing of learning initial states and learning driving maneuvers leads to a better evaluation using the cost function.

8. The method according to claim 1, wherein a computer program containing machine-readable instructions is executed on one or more computers to cause the one or more computers to carry out the method.

9. The method according to claim 8, wherein the computer program is stored on a machine-readable data carrier and/or download product.

10. The method according to claim 1, wherein a computer is configured to execute a computer program to perform the method.

Description

EMBODIMENT EXAMPLES

The Figures Shows:

[0036] FIG. 1 embodiment example of the method 100 for training a behavior planner 3;

[0037] FIG. 2 embodiment example of the method 200 comprising the complete chain of action all the way to the control of the vehicle 2;

[0038] FIG. 3 embodiment example of the method 300 for training a dynamics model 7.

[0039] FIG. 1 is a schematic flowchart of an embodiment example of the method 100 for training a behavior planner 3 for an at least partially self-driving target vehicle 2. Observation data 4 relating to the kinematics and/or dynamics that have been recorded during at least one test drive in a test vehicle 1 are used for the training.

[0040] In Step 110, at least one driving maneuver 5 that moves the test vehicle 1 from an initial state 5a to an end state 5b is identified from the observation data 4. According to Block 111, the driving maneuver 5, for example, can in particular be identified using a trained classifier model. This classifier model maps a time series of observation data 4 to at least one driving maneuver 5 that is consistent with that time series.

[0041] In Step 120, the maneuver end time 6, at which the test vehicle 1 reaches the end state 5b, is determined. In Step 130, a maneuver duration 8 required by the target vehicle 2 to perform the identified driving maneuver 5 is retrieved from a dynamics model 7 of the target vehicle 2. According to Block 131, for example, a dynamics model 7 of the target vehicle 2 can in particular be selected, which is configured to map a combination of an initial state 5a of the target vehicle 2 and a specification of a driving maneuver 5 to a maneuver duration 8.

[0042] In Step 140, observation data 4 from a time interval, the end of which is defined by the maneuver end time 6 and the start of which is before the maneuver end time 6 by the maneuver duration 8, are labeled with the identified driving maneuver 5. In Step 150, the behavior planner 3 is trained, using the labeled observation data 4, to map observation data 4 that indicate a state of the target vehicle 2 to at least one driving maneuver 5 to be carried out.

[0043] FIG. 2 is a schematic flowchart of an embodiment example of the method 200 comprising the complete chain of action all the way to the control of the at least partially self-driving vehicle 2.

[0044] In Step 210, a behavior planner 3 for the vehicle 2 is trained using the above-described method (100) according to any one of claims 1 to 5. In Step 220, observation data 4 relating to the kinematics and/or dynamics of the vehicle 2 are acquired by means of at least one sensor 2a of the vehicle 2.

[0045] In Step 230, the observation data 4 are mapped by means of the behavior planner 3 to at least one driving maneuver 5 to be carried out and, in Step 240, at least one actuator 2b acting on the driving dynamics of the vehicle 2 is controlled such that the vehicle 2 carries out the driving maneuver 5.

[0046] FIG. 3 is a schematic flowchart of an embodiment example of the method 300 for training a dynamics model 7 of a target vehicle 2.

[0047] In Step 310, learning initial states 5a and learning driving maneuvers 5 are provided. In Step 320, based on the learning initial states 5a, the learning driving maneuvers 5 are carried out on the target vehicle 2 and/or on a simulation model of the target vehicle 2. In Step 330, the respective time required to carry out the learning driving maneuvers 5 is determined and defined as the learning maneuver duration 8.

[0048] In Step 340, the learning initial states 5a and learning driving maneuvers 5 are fed to the dynamics model 7 and mapped by the dynamics model 7 to a respective maneuver duration 8. In Step 350, the agreement of the maneuver duration 8 with the learning maneuver duration 8 associated with the respective learning initial state 5a and learning driving maneuver 5 is evaluated using a specified cost function 9. In Step 360, the parameters 7a characterizing the behavior of the dynamics model 7 are optimized with the objective that the further processing of learning initial states 5a and learning driving maneuvers 5 leads to a better evaluation 9a by means of the cost function 9. The fully trained state of the parameters 7a is labeled with the reference sign 7a*.