Lane Segment Labeling for Relevant Agent Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing autonomous vehicle systems face challenges in generating relevant candidate future trajectories due to computational limitations, leading to error-prone predictions and limited trajectory evaluation, especially when agents deviate from lanes or perform maneuvers like U-turns, which complicates safe navigation and decision-making.
Innovation Solution
A machine learning model is trained to predict the likelihood of lane segments being taken by a target agent, allowing the on-board system to select only the highest-ranked lanes for generating path hypotheses and trajectories, thereby reducing the number of candidate trajectories while ensuring relevance and meeting computational requirements.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If heuristics are used to identify seed lane segments, then the system can generate candidate trajectories, but the approach is error-prone when agents deviate from lanes or perform maneuvers like U-turns
Solution Approach 1:
The system performs preliminary identification of relevant lane segments using a machine learning model before generating candidate trajectories. The model predicts which lane segments are likely to be taken by the target agent, creating a filtered set of seed lane segments that improves both accuracy and handles complex maneuvers better than traditional heuristics
Solution Approach 2:
The patent replaces traditional heuristic-based mechanical rules with a machine learning model that learns patterns from data. This substitution enables the system to automatically adapt to various driving scenarios including lane deviations and U-turns without requiring explicit programming for each case
2Productivity
If all lane segments within a larger radius are used as seed lane segments, then more candidate trajectories can be generated to address computational limits, but the number of trajectories becomes too large to evaluate timely
Solution Approach 1:
The system extracts only the most relevant lane segments from the complete set of available lane segments using a machine learning model. By taking out only the high-probability segments rather than processing all segments within a radius, the system reduces the number of candidate trajectories to a manageable size while maintaining quality
Solution Approach 2:
The patent changes the parameter selection criterion from a fixed spatial radius to a probability-based ranking. Instead of including all lane segments within a certain distance, the system uses the model's predicted probability to select only the top-ranked segments, optimizing the balance between trajectory coverage and computational efficiency
Data Source
AI summary
Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for generating candidate future trajectories for agents. One of the methods includes obtaining scene data characterizing a scene in an environment at a current time point; for each of a plurality of lane segments, processing a model input comprising (i) features of the lane segment and (ii) features of the target agent using a machine learning model that is configured to process the model input to generate a respective score for the lane segment that represents a likelihood that the lane segment will be a first lane segment traversed by the target agent after the current time point; selecting, as a set of seed lane segments, a proper subset of the plurality of lane segments based on the respective scores; and generating a plurality of candidate future trajectories for the target agent.