Kai Olav Ellefsen

CRRL: A Causality-Based Reinforcement Learning Framework for Autonomous System Recovery

Safia Fatima, Kai Olav Ellefsen, Leon Moonen

Traditional reinforcement learning (RL) for recovery in autonomous systems lacks causal understanding and generalizes poorly to novel failure scenarios. RL policies often stall in failure states, spending up to 70% of an episode immobilized. Rule-based recovery alone is inadequat…Read more