Pareto-Dominance Cost Learning for Robust AV Motion Planning
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing machine-learning techniques for autonomous vehicles are susceptible to suboptimal demonstrations, leading to inefficiencies and noise in motion planning, which can result in suboptimal performance and increased uncertainty.
Innovation Solution
The use of Pareto dominance-based learning methods to optimize the cost function of autonomous vehicles by minimizing subdominance, ensuring that the learned behavior dominates human demonstrations by a margin, thereby reducing the influence of outlier data and improving computational efficiency and accuracy.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If conventional cost function learning methods are used to minimize demonstration costs, then the system can learn from human demonstrations, but the system becomes highly sensitive to suboptimal outliers and demonstration noise
Solution Approach 1:
The patent introduces Pareto dominance as an intermediary criterion between demonstration data and cost function learning. Instead of directly minimizing demonstration costs (which is sensitive to outliers), the system uses Pareto dominance relationships as a mediator to filter and weight training examples, thereby reducing the influence of suboptimal demonstrations while preserving learning from valid human behavior
Solution Approach 2:
The patent changes the optimization parameter from direct cost minimization to Pareto dominance maximization. By transforming the learning objective from minimizing scalar costs to maximizing the number of Pareto-dominated demonstrations, the system fundamentally alters the parameter space to be more robust against outliers while maintaining learning effectiveness
2Adaptability or versatility
If maximum margin planning or maximum entropy inverse reinforcement learning is used, then the system can learn cost functions from demonstrations, but the system becomes highly sensitive to suboptimal outliers requiring careful data filtering
Solution Approach 1:
The patent enables the learning system to automatically filter suboptimal demonstrations through self-service mechanisms. By computing Pareto dominance relationships among all demonstrations, the system autonomously identifies and down-weights suboptimal examples without requiring manual data cleaning or expert intervention, thereby reducing operational complexity
Solution Approach 2:
The patent implements feedback loops where the learned cost function continuously evaluates demonstrations against current performance standards. Pareto dominance computation provides feedback about which demonstrations are suboptimal relative to others, automatically adjusting training weights based on this feedback to improve robustness without manual filtering
3Productivity
If conventional optimization methods are used for cost function learning, then the system can determine motion plans, but the system produces suboptimal performance with lateral nudging and jerk effects
Solution Approach 1:
The patent applies preliminary action by pre-computing Pareto dominance relationships and identifying support vectors before final cost function optimization. This preliminary filtering of suboptimal demonstrations based on dominance criteria prepares cleaner training data that leads to smoother motion plans with reduced lateral nudging and jerk effects in the final system
Data Source
AI summary
Techniques for improving the performance of an autonomous vehicle (AV) are described herein. A system can determine a plan for the AV in a driving scenario that optimizes an initial cost function of a control algorithm of the AV. The system can obtain data describing an observed human driving path in the driving scenario. Additionally, the system can determine for each cost dimension in the plurality of cost dimensions, a quantity that compares the estimated cost to the observed cost of the observed human driving path. Moreover, the system can determine a function of a sum of the quantities determined for each cost dimension in the plurality of cost dimensions. Subsequently, the system can use an optimization algorithm to adjust one or more weights of the plurality of weights applied to the plurality of cost dimensions to optimize the function of the sum of the quantities.