Discrete decision architecture for motion planning system of an autonomous vehicle
The present disclosure provides autonomous vehicle systems and methods that include or otherwise leverage a motion planning system that generates constraints as part of determining a motion plan for an autonomous vehicle (AV). In particular, a scenario generator within a motion planning system can generate constraints based on where objects of interest are predicted to be relative to an autonomous vehicle. A constraint solver can identify navigation decisions for each of the constraints that provide a consistent solution across all constraints. The solution provided by the constraint solver can be in the form of a trajectory path determined relative to constraint areas for all objects of interest. The trajectory path represents a set of navigation decisions such that a navigation decision relative to one constraint doesn't sacrifice an ability to satisfy a different navigation decision relative to one or more other constraints.
1. A method for controlling an autonomous vehicle, comprising:
(a) obtaining map data descriptive of an environment of the autonomous vehicle;
(b) obtaining sensor data descriptive of one or more actors in the environment of the autonomous vehicle;
(c) determining, based at least in part on the map data and the sensor data, a subset of the one or more actors that are relevant to the autonomous vehicle;
(d) determining discrete decisions with respect to actors in the subset of the one or more actors, wherein a respective discrete decision is determined from a group comprising a yield; a clear, and an ignore relative to a respective actor in the subset of the one or more actors;
(e) determining a trajectory for the autonomous vehicle that (i) is consistent with the discrete decisions with respect to the actors in the subset of the one or more actors and (ii) minimizes one or more cost functions, wherein determining the trajectory for the autonomous vehicle comprises minimizing a total cost of the one or more cost functions over a motion planning space, the motion planning space comprising a series of vehicle states, wherein determining the trajectory for the autonomous vehicle that is consistent with the discrete decisions comprises:
(i) generating a constraint relative to the respective discrete decision and the respective actor in the subset of the one or more actors;
(ii) determining a constraint area associated with the constraint, the constraint area defined relative to a multi-dimensional space comprising at least a time dimension and a distance dimension; and
(iii) determining the trajectory for the autonomous vehicle based on the constraint area associated with the constraint; and
(f) controlling the autonomous vehicle according to the trajectory.
2. The method of claim 1 , wherein (c) comprises determining that the actors in the subset of the one or more actors are interacting with a given lane associated, with the autonomous vehicle.
3. The method of claim 1 , wherein the respective discrete decision is determined from the group further comprising a lead and a follow relative to the respective actor in the subset of the one or more actors.
4. The method of claim 1 , wherein the respective discrete decision is determined from the group further comprising a nudge relative to the respective actor in the subset of the one or more actors.
5. The method of claim 1 , wherein (e) comprises determining a trajectory that (iii) is optimized relative to a lateral dimension associated with steering of the autonomous vehicle and a longitudinal dimension associated with at least one of braking or acceleration of the autonomous vehicle.
6. The method of claim 1 , wherein the one or more cost functions evaluate at least one of a dynamics cost, a buffer cost, a constraint violation cost, a blind spot cost, or an achieved performance objective reward.
7. The method of claim 1 , wherein (e) comprises: (1) determining a plurality of route maneuvers for the autonomous vehicle, (2) determining a best trajectory for respective route maneuvers in the plurality of route maneuvers, and (3) selecting a best route maneuver and trajectory pair.
8. The method of claim 1 , wherein (c) comprises processing the map data and the sensor data with one or more machine-learned models trained to generate the subset of the one or more actors that are relevant to the autonomous vehicle.
9. The method of claim 1 , wherein (d) comprises processing state data associated with the actors in the subset of the one or more actors with one or more machine-learned models trained to determine the discrete decisions.
10. The method of claim 1 , wherein (e) comprises processing the discrete decisions and the one or more cost functions with a machine-learned model trained to determine the trajectory.
11. The method of claim 1 , wherein comprises implementing an iterative linear quadratic regulator.
12. An autonomous vehicle control system for controlling an autonomous vehicle, the autonomous vehicle control system comprising:
one or more processors; and
one or more tangible, non-transitory, computer-readable media that store instructions that are executable to cause the autonomous vehicle control system to perform operations, the operations comprising:
(a) obtaining map data descriptive of an environment of the autonomous vehicle;
(b) obtaining sensor data descriptive of one or more actors in the environment of the autonomous vehicle;
(c) determining, based at least in part on the map data and the sensor data, a subset of the one or more actors that are relevant to the autonomous vehicle;
(d) determining discrete decisions with respect to the actors in the subset of the one or more actors, wherein a respective discrete decision is determined from a group comprising a yield, a clear, and an ignore relative to a respective actor in the subset of the one or more actors;
(e) determining a trajectory for the autonomous vehicle that (i) is consistent with the discrete decisions with respect to the actors in the subset of the one or more actors and (ii) minimizes one or more cost functions, wherein determining the trajectory for the autonomous vehicle comprises minimizing a total cost of the one or more cost functions over a motion planning space, the motion planning space comprising a series of vehicle states, wherein determining the trajectory for the autonomous vehicle that is consistent with the discrete decisions comprises:
(i) generating a constraint relative to the respective discrete decision and the respective actor in the subset of the one or more actors;
(ii) determining a constraint area associated with the constraint, the constraint area defined relative to a multi-dimensional space comprising at least a time dimension and a distance dimension; and
(iii) determining the trajectory for the autonomous vehicle based on the constraint area associated with the constraint; and
(f) controlling the autonomous vehicle according to the trajectory.
13. The autonomous vehicle control system of claim 12 , wherein (c) comprises determining that the subset of the one or more actors are interacting with a given lane associated with the autonomous vehicle.
14. The autonomous vehicle control system of claim 12 , wherein the respective discrete decision is determined from the group further comprising a lead and a follow relative to the respective actor in the subset of the one or more actors.
15. The autonomous vehicle control system of claim 12 , wherein the respective discrete decision is determined from the group further comprising a nudge relative to the respective actor in the subset of the one or more actors.
16. The autonomous vehicle control system of claim 12 , wherein (e) comprises determining a trajectory that (iii) is optimized relative to a lateral dimension associated with steering of the autonomous vehicle and a longitudinal dimension associated with at least one of braking or acceleration of the autonomous vehicle.
17. The autonomous vehicle control system of claim 12 , wherein the one or more cost functions evaluate at least one of a dynamics cost, a buffer cost, a constraint violation cost, a blind spot cost, or an achieved performance objective reward.
18. The autonomous vehicle control system of claim 12 , wherein (e) comprises: (1) determining a plurality of route maneuvers for the autonomous vehicle, (2) determining a best trajectory for respective route maneuvers in the plurality of route maneuvers, and (3) selecting a best route maneuver and trajectory pair.
19. The autonomous vehicle control system of claim 12 , wherein (e) comprises implementing an iterative linear quadratic regulator.
20. An autonomous vehicle; comprising:
one or more processors; and
one or more tangible, non-transitory, computer-readable media that store instructions that are executable to cause the one or more processors to perform operations, the operations comprising:
(a) obtaining map data descriptive of an environment of the autonomous vehicle;
(b) obtaining sensor data descriptive of one or more actors in the environment of the autonomous vehicle;
(c) determining, based at least in part on the map data and the sensor data, a subset of the one or more actors that are relevant to the autonomous vehicle;
(d) determining discrete decisions with respect to the actors in the subset of the one or more actors, wherein a respective discrete decision is determined from a group comprising a yield, a clear, and an ignore relative to a respective actor in the subset of the one or more actors;
(e) determining a trajectory for the autonomous vehicle that (i) is consistent with the discrete decisions with respect to the actors in the subset of the one or more actors and (ii) minimizes one or more cost functions, wherein determining the trajectory for the autonomous vehicle comprises minimizing a total cost of the one or more cost functions over a motion planning space, the motion planning space comprising a series of vehicle states, wherein determining the trajectory for the autonomous vehicle that is consistent with the discrete decisions comprises:
(i) generating a constraint relative to the respective discrete decision and the respective actor in the subset of the one or more actors;
(ii) determining a constraint area associated with the constraint, the constraint area defined relative to a multi-dimensional space comprising at least a time dimension and a distance dimension; and
(iii) determining the trajectory for the autonomous vehicle based on the constraint area associated with the constraint; and
(f) controlling the autonomous vehicle according to the trajectory.