Combating Reinforcement Learning's Sisyphean Curse With Intrinsic Fear

Abstract

Many practical environments contain catastrophic states that an optimal agent would visit infrequently or never. Even on toy problems, Deep Reinforcement Learning (DRL) agents tend to periodically revisit these states upon forgetting their existence under a new policy. We introduce intrinsic fear (IF), a learned reward shaping that guards DRL agents against periodic catastrophes. IF agents possess a fear model trained to predict the probability of imminent catastrophe. This score is then used to penalize the Q-learning objective. Our theoretical analysis bounds the reduction in average return due to learning on the perturbed objective. We also prove robustness to classification errors. As a bonus, IF models tend to learn faster, owing to reward shaping. Experiments demonstrate that intrinsic-fear DQNs solve otherwise pathological environments and improve on several Atari games.

Stats

• citations0
• S2 citations—
• github stars0
• HF likes0
• heat score0.00
• arxiv keylipton2016combating

Related papers

• Regret-based Defense In Adversarial Reinforcement Learning (2023)0.00
• Training A Resilient Q-network Against Observational Interference (2021)0.00
• Towards Safe Reinforcement Learning Via Constraining Conditional Value-at-risk (2022)0.00
• DOPE: Doubly Optimistic And Pessimistic Exploration For Safe Reinforcement Learning (2021)0.00
• An Information-theoretic Optimality Principle For Deep Reinforcement Learning (2017)0.00
• Disturbing Reinforcement Learning Agents With Corrupted Rewards (2021)0.00
• Real-time Adversarial Perturbations Against Deep Reinforcement Learning Policies: Attacks And Defenses (2021)0.00
• Deep Reinforcement Learning With Decorrelation (2019)0.00