Autonomous vehicles (AVs) require rigorous testing before deployment. Due to the complexity of these systems, formal verification may be impossible and real-world testing may be dangerous and expensive during development.
This is where AST fits in. Anthony Corso presents his work on Adaptive Stress Testing (AST), a technique for automatically finding the most likely failures of an autonomous system in simulation. AST treats the AV as a black box and uses reinforcement learning to manipulate the driving environment toward challenging scenarios.
He will demonstrate the discovery of failures for aircraft collision avoidance systems, simple autonomous vehicles, and a vision-based controller that uses a neural network.
Get the slides here: