New method aims to improve safety of AI-based perception systems

VERA Brief

AI-generated. Grounded in the article and its cited sources.

Researchers have developed a two-step strategy to improve the safety of AI-based perception systems in autonomous vehicles. This method uses offline computation to identify uncertainties and a runtime risk heuristic to guide control decisions, aiming to prevent unsafe actions.

Key facts

• A new strategy addresses safety concerns in learning-based perception for autonomous systems.
• The approach uses offline computation to characterize uncertainties and a runtime risk heuristic to guide control decisions.
• The strategy aims to prevent unsafe actions by selecting specific states from uncertain estimates for control decisions.
• In simulations, the new method preserved safety in 73% of scenarios where the original system had safety violations.
• The strategy reportedly does not introduce excessive conservatism, with a small average increase in completion time.