Autonomous Vehicle Control Training From Human Driving Deviations
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in accurately interpreting their surroundings and planning motion to navigate complex environments, as existing systems struggle to replicate human driving behavior consistently, requiring continuous adjustment of numerous cost functions to account for various constraints.
Innovation Solution
The training of cost and value functions for autonomous vehicle control systems using machine learning models, such as decision trees and neural networks, incorporating manual driving data to identify and update deviations between predicted and actual vehicle trajectories, allowing the system to better mimic human driving behavior.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If continuous adjustment of numerous cost functions is performed to account for various constraints, then the autonomous vehicle can navigate complex environments more effectively, but the system complexity and computational burden increase significantly
Solution Approach 1:
The patent combines multiple cost functions into a unified machine learning model that learns optimal navigation strategies from human driving data. Instead of manually adjusting numerous separate cost functions, the system merges them into a single trained model that outputs navigation commands directly, reducing system complexity while maintaining adaptability to complex environments
Solution Approach 2:
The patent transforms the cost function parameters from manually tuned values to learned parameters from training data. By changing how these parameters are determined (from manual adjustment to machine learning), the system reduces the burden of continuous parameter tuning while improving performance in complex environments
2Reliability
If machine learning models are used to represent cost functions and learn from manual driving data, then the autonomous vehicle can better replicate human driving behavior, but the training and computation time increase
Solution Approach 1:
The patent performs preliminary training of the machine learning model using extensive manual driving data before deployment. By conducting this training action in advance (offline), the system achieves high reliability in replicating human driving behavior without incurring training time delays during actual operation
Solution Approach 2:
The patent uses copying by training the model on recorded human driving data to replicate human behavior patterns. This allows the system to learn from extensive training data without requiring real-time training, as the model copies human decision-making patterns from the training examples
Data Source
AI summary
Techniques are disclosed for training one or more cost functions of an autonomous vehicle (“AV”) control system based on difference between data generated using the AV control system and manual driving data. In many implementations, manual driving data captures action(s) of a vehicle controlled by a manual driver. Additionally or alternatively, multiple AV control systems can be evaluated by comparing deviations for each AV control system, where the deviations are determined using the same set of manual driving data.