Trajectory Feature Planning for Autonomous Vehicle Path Ranking
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Training autonomous vehicles to accurately perceive their environment is complicated, and evaluating and selecting trajectories is difficult due to compute resource constraints and the complexity of learning reward functions in state-action spaces.
Innovation Solution
A training environment is used to train a machine learning model to learn an interpretable reward function in a trajectory space by providing sets of trajectory features, allowing the model to score and rank trajectories based on these features.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If traditional state-action space methods are used to train autonomous vehicles, then the system can learn from detailed state transitions, but the compute resources required become excessively large and training complexity increases
Solution Approach 1:
The patent extracts only the essential trajectory features (position, velocity, acceleration, heading, curvature) from the complex state-action space, separating the critical motion parameters from the full environmental state. This extraction reduces the training dimensionality while preserving the information needed for accurate trajectory evaluation.
Solution Approach 2:
The trajectory representation is segmented into discrete feature components (position, velocity, acceleration, heading, curvature) that can be independently processed and evaluated. This segmentation allows the system to handle trajectory data in manageable units rather than as a monolithic state vector.
2Reliability
If detailed state-action space learning is used, then the model can capture complex trajectory dynamics, but the training time and computational resources required increase significantly
Solution Approach 1:
The patent changes the parameter representation from full state-action vectors to a simplified set of trajectory features. By transforming the state representation into position, velocity, acceleration, heading, and curvature parameters, the system achieves reliable trajectory prediction with reduced computational requirements and faster training time.
3Measurement precision
If multiple types of images are used for environment perception, then the perception accuracy improves, but the training complexity and data processing requirements increase
Solution Approach 1:
The trajectory feature extraction system serves multiple functions: it processes data from various image sources, predicts object trajectories, evaluates planning options, and provides feedback for training. This multi-functional approach consolidates several processing streams into a unified feature representation system.
Data Source
AI summary
A trajectory planning system can be used to select a trajectory for an autonomous vehicle. The trajectory planning system may generate multiple trajectories and extract features from the generated trajectories. The trajectory planning system may evaluate the trajectories based on the extracted features and select a trajectory for the vehicle based on the evaluation. The selected trajectory may be used to control the vehicle.