Persisting Predicted Objects for Robustness to Perception Issues in Autonomous Driving

A system comprising:

• one or more sensors coupled to an autonomous vehicle and configured to detect one or more objects in an environment of the autonomous vehicle;
• and
• a computing device comprising a processor and memory, wherein the memory includes instructions that, when executed by the processor, will cause the processor to:
• predict a first set of predicted object trajectories comprising one or more trajectories for each of the detected one or more objects,
• use a motion planning model to generate a plurality of candidate autonomous vehicle trajectories for the autonomous vehicle, wherein at least one predicted object trajectory from a persisted prediction cache is used as an input to the motion planning model for each of the candidate autonomous vehicle trajectories,
• score each of the candidate autonomous vehicle trajectories according to a cost function,
• use the scoring to select a final autonomous vehicle trajectory for execution,
• determine which of the predicted object trajectories affected the final autonomous vehicle trajectory and which did not do so,
• add the predicted object trajectories that affected the final autonomous vehicle trajectory to the persisted prediction cache,
• exclude from the persisted prediction cache any predicted object trajectories that did not affect the final autonomous vehicle trajectory, and
• using a motion planning system of the autonomous vehicle, execute the final autonomous vehicle trajectory to cause the autonomous vehicle to move along the final autonomous vehicle trajectory.

A computer-readable medium comprising programming instructions that are configured to cause a processor to:

• receive information relating to one or more objects that one or more sensors coupled to an autonomous vehicle detected in an environment of the autonomous vehicle;
• predict a first set of predicted object trajectories comprising one or more trajectories for each of the detected one or more objects;
• use a motion planning model to generate a plurality of candidate autonomous vehicle trajectories for the autonomous vehicle, wherein at least one predicted object trajectory from a persisted prediction cache is used as an input to the motion planning model for each of the candidate autonomous vehicle trajectories;
• score each of the candidate autonomous vehicle trajectories according to a cost function;
• use the scoring to select a final autonomous vehicle trajectory for execution;
• determine which of the predicted object trajectories affected the final autonomous vehicle trajectory and which did not do so;
• add the predicted object trajectories that affected the final autonomous vehicle trajectory to the persisted prediction cache;
• exclude from the persisted prediction cache any predicted object trajectories that did not affect the final autonomous vehicle trajectory;
• and
• using a motion planning system of the autonomous vehicle, execute the final autonomous vehicle trajectory to cause the autonomous vehicle to move along the final autonomous vehicle trajectory.