Interpretable Motion Planner Using Cost Volumes for Autonomous Vehicles
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current motion planning systems for autonomous vehicles are inefficient in generating accurate and interpretable motion plans, particularly in complex urban scenarios, due to their reliance on divided subtasks like perception, prediction, and control, which lack end-to-end learnability and interpretability.
Innovation Solution
A machine-learned motion planning model that integrates sensor data and map data to generate a cost volume representing the desirability of future locations, enabling the selection of optimal trajectories through a backbone network and trajectory generator, while providing interpretable intermediate representations for object detection and motion prediction.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Ease of manufacture
If motion planning systems use divided subtasks (perception, prediction, control), then the system structure is modular and easier to implement, but the system lacks end-to-end learnability and interpretability
Solution Approach 1:
The patent combines multiple divided subtasks (perception, prediction, motion planning) into a unified end-to-end neural network model. The network processes sensor data and map data through integrated layers that simultaneously perform object detection, motion prediction, and trajectory generation, eliminating the need for separate modular components and enabling true end-to-end learning.
Solution Approach 2:
The neural network model serves multiple functions within a single architecture: it performs perception (object detection), prediction (motion forecasting), and control (trajectory planning) simultaneously. This multi-functional approach allows the system to learn all tasks jointly through end-to-end training, improving both learnability and interpretability while maintaining implementation feasibility.
2Ease of operation
If motion planning systems use divided subtasks, then the system is easier to interpret individually, but the overall system lacks interpretability and accuracy in complex scenarios
Solution Approach 1:
The patent introduces intermediate representation layers within the neural network that provide interpretability at each processing stage. These intermediate layers represent perceptual features, predicted motions, and planning decisions in a structured manner that can be analyzed, while the overall end-to-end architecture maintains high accuracy by learning optimal transformations between these representations and final trajectories.
3Measurement precision
If the system processes sensor data and map data through multiple layers, then the model can generate accurate trajectories, but the computational complexity increases
Solution Approach 1:
The neural network is organized into distinct functional layers (perception layers, prediction layers, planning layers) that process data in a hierarchical manner. Each layer focuses on specific aspects of the problem, breaking down the complex computation into manageable segments that can be optimized independently while maintaining overall trajectory accuracy.
Data Source
AI summary
Systems and methods for generating motion plans including target trajectories for autonomous vehicles are provided. An autonomous vehicle may include or access a machine-learned motion planning model including a backbone network configured to generate a cost volume including data indicative of a cost associated with future locations of the autonomous vehicle. The cost volume can be generated from raw sensor data as part of motion planning for the autonomous vehicle. The backbone network can generate intermediate representations associated with object detections and objection predictions. The motion planning model can include a trajectory generator configured to evaluate one or more potential trajectories for the autonomous vehicle and to select a target trajectory based at least in part on the cost volume generate by the backbone network.