Towards Safer Self-Driving Through Great PAIN (Physically Adversarial Intelligent Networks)
This work addresses safety issues in self-driving vehicles by generating adversarial scenarios to enhance training, though it is incremental as it builds on existing adversarial methods in simulation.
The paper tackled the problem of improving self-driving vehicle safety by addressing overfit and poor generalizability in neural networks due to limited data. It introduced a Physically Adversarial Intelligent Network (PAIN) in simulation, resulting in a protagonist agent that became more resilient to environmental uncertainty and less prone to corner case failures, with increased mean-time-to-failure and distance traveled under non-hostile conditions.
Automated vehicles' neural networks suffer from overfit, poor generalizability, and untrained edge cases due to limited data availability. Researchers synthesize randomized edge-case scenarios to assist in the training process, though simulation introduces potential for overfit to latent rules and features. Automating worst-case scenario generation could yield informative data for improving self driving. To this end, we introduce a "Physically Adversarial Intelligent Network" (PAIN), wherein self-driving vehicles interact aggressively in the CARLA simulation environment. We train two agents, a protagonist and an adversary, using dueling double deep Q networks (DDDQNs) with prioritized experience replay. The coupled networks alternately seek-to-collide and to avoid collisions such that the "defensive" avoidance algorithm increases the mean-time-to-failure and distance traveled under non-hostile operating conditions. The trained protagonist becomes more resilient to environmental uncertainty and less prone to corner case failures resulting in collisions than the agent trained without an adversary.