Autonomy Cognizant Agent for AV Decision Making
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Autonomous vehicles face challenges in determining the appropriate level of operational autonomy, often relying on human intervention due to limited sensor data availability and occlusions, which can lead to inefficient operation and increased reliance on human assistance.
Innovation Solution
An autonomy cognizant agent (ACA) is introduced, which uses an extended Stochastic Shortest Path problem to calculate a policy that maps environment states and autonomy levels to actions, allowing it to generate plans across multiple autonomy levels and proactively manage human feedback, thereby reducing reliance on human intervention.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If the autonomous vehicle relies on human intervention due to limited sensor data availability, then the vehicle can handle uncertain situations, but the operational efficiency decreases and human assistance is required more frequently
Solution Approach 1:
The system implements a feedback model that learns from human feedback signals (approval, correction, takeover) to improve the autonomy model. The human feedback serves as training data that enables the system to better predict when human intervention will be needed, reducing unnecessary interventions while maintaining reliability in uncertain situations
Solution Approach 2:
The system proactively requests human feedback before executing actions when uncertainty is detected, rather than waiting for failures or errors to occur. This preliminary action allows the system to learn from human decisions in advance, improving future autonomy decisions and reducing operational delays
2Reliability
If the autonomous vehicle requests human feedback frequently to ensure safe operation, then the reliability improves, but the time consumption increases
Solution Approach 1:
The system dynamically adjusts the threshold for requesting human feedback based on learned patterns from the feedback model. As the system accumulates experience, it refines its understanding of when human intervention is truly necessary, optimizing the balance between safety and time efficiency by changing the parameters that trigger feedback requests
3Productivity
If the autonomous vehicle operates with higher autonomy levels, then the productivity increases, but the difficulty of detecting and measuring the appropriate autonomy level increases
Solution Approach 1:
The system replaces manual, mechanical judgment of autonomy levels with an automated autonomy model that calculates appropriate autonomy levels based on learned patterns from human feedback. This substitution enables objective, consistent measurement of autonomy appropriateness without relying on subjective human assessment
Solution Approach 2:
The system autonomously determines its own appropriate autonomy level by processing sensor data through the autonomy model and feedback model, without requiring external human evaluation. This self-service capability enables continuous optimization of autonomy level selection based on accumulated experience
Data Source
AI summary
A first method includes detecting, based on sensor data, an environment state; selecting an action based on the environment state; determining an autonomy level associated with the environment state and the action; and performing the action according to the autonomy level. The autonomy level can be selected based at least on an autonomy model and a feedback model. A second method includes calculating, by solving an extended Stochastic Shortest Path (SSP) problem, a policy for solving a task. The policy can map environment states and autonomy levels to actions and autonomy levels. Calculating the policy can include generating plans that operate across multiple levels of autonomy.