Prediction-based system and method for trajectory planning of autonomous vehicles
A prediction-based system and method for trajectory planning of autonomous vehicles are disclosed. A particular embodiment is configured to: receive training data and ground truth data from a training data collection system, the training data including perception data and context data corresponding to human driving behaviors; perform a training phase for training a trajectory prediction module using the training data; receive perception data associated with a host vehicle; and perform an operational phase for extracting host vehicle feature data and proximate vehicle context data from the perception data, generating a proposed trajectory for the host vehicle, using the trained trajectory prediction module to generate predicted trajectories for each of one or more proximate vehicles near the host vehicle based on the proposed host vehicle trajectory, determining if the proposed trajectory for the host vehicle will conflict with any of the predicted trajectories of the proximate vehicles, and modifying the proposed trajectory for the host vehicle until conflicts are eliminated.
1. A system comprising:
a data processor; and
a prediction-based trajectory planning module, executable by the data processor, the prediction-based trajectory planning module being configured to:
receive perception data associated with a host vehicle;
extract host vehicle feature data and proximate vehicle context data from the perception data;
generate a proposed trajectory for the host vehicle;
use a trained trajectory prediction module to generate predicted trajectories for each of one or more proximate vehicles, the trained trajectory prediction module having been trained using training data comprising perception data and context data corresponding to human driving behaviors, the predicted trajectories for each of the one or more proximate vehicles corresponding to likely actions or reactions by each of the one or more proximate vehicles based on the human driving behavior context data if the host vehicle follows the proposed host vehicle trajectory; and
modify the proposed trajectory for the host vehicle if the proposed trajectory for the host vehicle will conflict with any of the predicted trajectories of the one or more proximate vehicles.
2. The system of claim 1 being further configured to filter and smooth the perception data.
3. The system of claim 1 being further configured to generate coordinate transformations of the perception data relative to the one or more proximate vehicles.
4. The system of claim 1 wherein the training data comprises labeling data that comprises context information defining directionality and rate behaviors of vehicles represented in the training data.
5. The system of claim 1 wherein the training data comprises labeling data that comprises context information defining directionality and rate behaviors of vehicles represented in the training data, the context data further defining a left turn, no turn, or a right turn.
6. The system of claim 1 wherein modifying the proposed trajectory for the host vehicle comprises:
determining if the proposed trajectory for the host vehicle will conflict with any of the predicted trajectories of the one or more proximate vehicles; and
modifying the proposed trajectory for the host vehicle based on the determined conflicts until the conflicts are eliminated.
7. The system of claim 1 being further configured to use regression to predict acceleration of a vehicle.
8. The system of claim 1 being further configured to determine if any of the predicted trajectories for the one or more proximate vehicles may cause the host vehicle to violate a pre-defined goal based on a related score being below a minimum acceptable threshold.
9. The system of claim 1 wherein the proposed trajectory for the host vehicle is output to a vehicle control subsystem causing the host vehicle to follow the output proposed trajectory.
10. A method comprising:
receiving perception data associated with a host vehicle;
extracting host vehicle feature data and proximate vehicle context data from the perception data;
generating a proposed trajectory for the host vehicle;
using a trained trajectory prediction module to generate predicted trajectories for each of one or more proximate vehicles, the trained trajectory prediction module having been trained using training data comprising perception data and context data corresponding to human driving behaviors, the predicted trajectories for each of the one or more proximate vehicles corresponding to likely actions or reactions by each of the one or more proximate vehicles based on the human driving behavior context data if the host vehicle follows the proposed host vehicle trajectory; and
modifying the proposed trajectory for the host vehicle if the proposed trajectory for the host vehicle will conflict with any of the predicted trajectories of the one or more proximate vehicles.
11. The method of claim 10 wherein the proximate vehicle context data comprises proximate vehicle position and proximate vehicle velocity.
12. The method of claim 10 comprising determining a position of each proximate vehicle relative to the host vehicle.
13. The method of claim 10 comprising obtaining perception data from an array of perception information gathering devices or sensors.
14. The method of claim 10 wherein the training data comprises labeling data obtained from human labelers or automated labeling processes.
15. The method of claim 10 wherein the perception data comprising data received from a sensor from the group consisting of: a camera or image capture device, a Global Positioning System (GPS) transceiver, and a laser range finder/LIDAR unit.
16. The method of claim 10 comprising predicting acceleration of a vehicle.
17. The method of claim 10 comprising determining if any of the predicted trajectories for the one or more proximate vehicles may cause the host vehicle to violate a pre-defined goal.
18. The method of claim 10 comprising causing the host vehicle to follow the proposed trajectory.
19. A non-transitory machine-readable storage medium embodying instructions which, when executed by a machine, cause the machine to:
receive perception data associated with a host vehicle;
extract host vehicle feature data and proximate vehicle context data from the perception data;
generate a proposed trajectory for the host vehicle;
use a trained trajectory prediction module to generate predicted trajectories for each of one or more proximate vehicles, the trained trajectory prediction module having been trained using training data comprising perception data and context data corresponding to human driving behaviors, the predicted trajectories for each of the one or more proximate vehicles corresponding to likely actions or reactions by each of the one or more proximate vehicles based on the human driving behavior context data if the host vehicle follows the proposed host vehicle trajectory; and
modify the proposed trajectory for the host vehicle if the proposed trajectory for the host vehicle will conflict with any of the predicted trajectories of the one or more proximate vehicles.
20. The non-transitory machine-readable storage medium of claim 19 being configured to generate predicted accelerations for each of the one or more proximate vehicles near the host vehicle.