Transfer Learning for Efficient Iterative Safety Validation

Safety validation is important during the development of safety-critical autonomous systems but can require significant computational effort. Existing algorithms often start from scratch each time system under test changes. We apply transfer learning to improve efficiency reinforcement based safety when applied related systems. Knowledge previous tasks encoded through action value function and transferred future with a learned set attention weights. Including state transformation for source task performance even have substantially different failure modes. conduct experiments on in gridworld driving scenarios. show that initial final reduce number training steps.