Autonomous Vehicle Motion Planning With Joint Behavior-Trajectory Learning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle motion planning systems lack an integrated approach that jointly optimizes behavioral planning and trajectory planning, leading to suboptimal decision-making in complex scenarios.
Innovation Solution
A machine-learned motion planning system with a unified cost function that jointly trains behavioral and trajectory planning stages using a combined loss function, enabling the generation of optimal target trajectories that balance safety, comfort, feasibility, and adherence to traffic rules.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If separate training approaches are used for behavioral planning and trajectory planning, then each stage can be developed independently, but the overall motion planning performance is suboptimal due to lack of joint optimization
Solution Approach 1:
The patent merges the behavioral planning stage and trajectory planning stage into a unified machine-learned motion planning system that is jointly trained using a combined loss function. This integration allows both planning stages to optimize their parameters simultaneously based on overall motion planning performance, resolving the contradiction by prioritizing system-level optimization over independent development simplicity.
Solution Approach 2:
The unified cost function serves multiple purposes simultaneously: it guides behavioral planning decisions, optimizes trajectory generation, and evaluates overall motion planning performance. This multi-functional approach allows a single training framework to improve both planning stages together, achieving reliable motion planning without requiring separate specialized training systems for each stage.
2Manufacturing precision
If iterative hand-tuning of planner costs is performed, then planning parameters can be optimized, but the process is time-consuming and requires extensive manual adjustment
Solution Approach 1:
The patent replaces the manual iterative hand-tuning process with an automated machine learning training system. The combined loss function automatically guides the optimization of planning parameters through gradient-based learning, substituting the mechanical process of manual adjustment with an automated computational optimization process that achieves similar or superior parameter precision without requiring human intervention.
Solution Approach 2:
The motion planning system performs self-optimization through the joint training process. The unified cost function automatically adjusts the parameters of both behavioral and trajectory planning stages based on performance feedback, enabling the system to optimize its own planning parameters without external manual tuning, thereby eliminating time-consuming hand-adjustment processes.
3Adaptability or versatility
If a unified cost function is used for joint training, then optimal target trajectories can be generated that balance multiple factors, but the training system becomes more complex
Solution Approach 1:
The unified cost function is designed to be multi-functional, simultaneously evaluating behavioral planning quality, trajectory feasibility, safety constraints, and comfort criteria. This single comprehensive function replaces multiple separate cost functions, achieving versatile trajectory optimization capability while actually simplifying the overall training architecture by consolidating multiple evaluation criteria into one unified framework.
Data Source
AI summary
Systems and methods for generating motion plans for autonomous vehicles are provided. An autonomous vehicle can include a machine-learned motion planning system including one or more machine-learned models configured to generate target trajectories for the autonomous vehicle. The model(s) include a behavioral planning stage configured to receive situational data based at least in part on the one or more outputs of the set of sensors and to generate behavioral planning data based at least in part on the situational data and a unified cost function. The model(s) includes a trajectory planning stage configured to receive the behavioral planning data from the behavioral planning stage and to generate target trajectory data for the autonomous vehicle based at least in part on the behavioral planning data and the unified cost function.