Interactive Motion Planning With Joint Trajectory Prediction
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current autonomous vehicle systems lack a comprehensive framework for jointly predicting the motion of surrounding objects and planning their own motion, leading to inefficiencies in navigating complex environments, particularly due to the lack of interdependence modeling in object interactions.
Innovation Solution
A machine-learned model framework that predicts candidate object trajectories and determines vehicle motion trajectories based on predicted interactions, using cost functions to score and select optimal vehicle motion paths that account for object interactions, enabling more accurate and efficient navigation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If autonomous vehicle systems use traditional separate prediction and planning modules, then system simplicity is maintained, but prediction accuracy and navigation efficiency deteriorate due to lack of interdependence modeling
Solution Approach 1:
The patent combines separate prediction and planning modules into a unified machine-learned model framework that jointly predicts object trajectories and plans vehicle motion. This integration allows the system to model interdependencies between object interactions and vehicle maneuvers, improving prediction accuracy by considering how objects will respond to planned actions rather than treating prediction and planning as independent sequential tasks.
2Productivity
If autonomous vehicle systems model interdependent object interactions comprehensively, then navigation efficiency improves, but processing time increases
Solution Approach 1:
The system performs preliminary action by pre-computing and storing interaction models during the training phase. The machine-learned model is trained offline on large datasets to learn object interaction patterns, so that during real-time operation, the system can quickly query pre-learned predictions rather than computing interactions from scratch. This shifts computational burden to the training phase, enabling efficient real-time navigation with comprehensive interaction modeling.
3Adaptability or versatility
If autonomous vehicle systems use simplified motion planning without interaction prediction, then processing speed is maintained, but ability to perform complex maneuvers deteriorates
Solution Approach 1:
The patent replaces traditional mechanical rule-based motion planning systems with a machine-learned model framework. Instead of using predefined rules and constraints for planning maneuvers, the system uses learned patterns from training data to predict object responses and generate appropriate motion plans. This substitution enables complex maneuvers by leveraging learned behavioral patterns rather than rigid rule-based systems, achieving both versatility and efficiency.
Data Source
AI summary
Example aspects of the present disclosure describe determining, using a machine-learned model framework, a motion trajectory for an autonomous platform. The motion trajectory can be determined based at least in part on a plurality of costs based at least in part on a distribution of probabilities determined conditioned on the motion trajectory.