Autonomous Vehicle Perception for Future Velocity Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Contemporary perception systems for autonomous platforms primarily output instantaneous velocities and accelerations, lacking the ability to predict future velocities and environmental context, which hinders advanced motion planning tasks such as merging and lane changing.
Innovation Solution
A perception system that integrates multiple sensor sweeps and a machine-learned model to predict future velocities and uncertainty scores, refining these predictions with an object tracker model to enhance motion planning accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Device complexity
If contemporary perception systems output only instantaneous velocity and acceleration, then the system complexity is reduced, but the motion planning capability for complex tasks such as merging and lane changing is insufficient
Solution Approach 1:
The perception system performs preliminary action by predicting future velocities and positions of actors before the motion planning stage. The machine-learned perception model generates predicted future velocities at multiple time steps ahead, providing advance information that enables better motion planning decisions for complex tasks like merging and lane changing.
Solution Approach 2:
The system transitions from outputting only instantaneous velocity and acceleration to outputting predicted future velocities at multiple time steps. This adds a temporal dimension to the perception output, transforming it from a single-point-in-time measurement to a time-series prediction that captures future motion states.
2Measurement precision
If predicted future velocities are generated by processing multiple sweeps of sensor data with machine-learned models, then the accuracy of motion planning is improved, but the computational processing time increases
Solution Approach 1:
The machine-learned perception model performs preliminary computation by predicting future velocities at multiple time steps in advance. This allows the motion planning system to receive pre-computed velocity predictions rather than calculating them in real-time during planning, reducing the critical path computation time.
Solution Approach 2:
The system processes multiple sweeps of sensor data continuously to generate smooth, continuous predictions of future velocities. This continuous processing approach maintains accurate predictions while optimizing computational efficiency by utilizing all available sensor data sweeps rather than processing them individually.
3Loss of information
If the perception model predicts future velocities at multiple time steps ahead, then the context awareness for motion planning is enhanced, but the uncertainty of predictions increases
Solution Approach 1:
The system performs preliminary prediction at multiple future time steps, providing the motion planning system with advance knowledge of expected actor positions and velocities. This allows planners to evaluate multiple future scenarios and select trajectories that maintain safe distances and appropriate speeds throughout the prediction horizon.
Solution Approach 2:
The system incorporates uncertainty estimates as feedback to the motion planning process. By providing confidence measures alongside predicted velocities, the planner can adjust its risk tolerance and select more conservative trajectories when prediction uncertainty is high, maintaining reliability despite extended prediction horizons.
Data Source
AI summary
An autonomous platform can obtain sensor data descriptive of an actor in an environment of an autonomous vehicle and at least a portion of the environment of the autonomous vehicle that does not include the actor, the sensor data comprising at least one sweep of the environment of the autonomous vehicle; process the sensor data with a machine-learned perception model to generate a detection of the actor and one or more predicted future velocities; and determine a motion trajectory for the autonomous vehicle based at least in part on the detection and the one or more predicted future velocities.