Summary of Dynamic Model Predictive Shielding For Provably Safe Reinforcement Learning, by Arko Banerjee et al.
Dynamic Model Predictive Shielding for Provably Safe Reinforcement Learning
by Arko Banerjee, Kia Rahmani, Joydeep Biswas, Isil Dillig
First submitted to arxiv on: 22 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: Machine Learning (cs.LG)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The proposed Dynamic Model Predictive Shielding (DMPS) approach optimizes reinforcement learning objectives while maintaining provable safety in complex tasks. Building upon the effective Model Predictive Shielding (MPS) method, DMPS employs a local planner to dynamically select safe recovery actions that balance short-term progress and long-term rewards. The planner utilizes the neural policy to estimate long-term rewards, allowing it to observe beyond its planning horizon. Conversely, the neural policy learns from the recovery plans proposed by the planner, converging to high-performing and safe policies in practice. This approach guarantees safety during and after training, with bounded recovery regret that decreases exponentially with planning horizon depth. DMPS outperforms several state-of-the-art baselines, achieving higher rewards while rarely requiring shield interventions. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary DMPS is a new way to make reinforcement learning safer and more effective. It uses a combination of two techniques: a local planner that makes decisions about what actions are safe, and a neural policy that learns from those decisions. The planner helps the neural policy by suggesting recovery plans when it would take a risk, while the neural policy provides feedback to the planner on how well its suggestions worked out. This approach ensures safety during training, and even after training is finished, there’s little chance of taking a risky action. The results show that DMPS performs better than other methods, with fewer mistakes and higher rewards. |
Keywords
» Artificial intelligence » Reinforcement learning
