Summary of Safe Offline Reinforcement Learning with Feasibility-guided Diffusion Model, by Yinan Zheng et al.
Safe Offline Reinforcement Learning with Feasibility-Guided Diffusion Model
by Yinan Zheng, Jianxiong Li, Dongjie Yu, Yujie Yang, Shengbo Eben Li, Xianyuan Zhan, Jingjing Liu
First submitted to arxiv on: 19 Jan 2024
Categories
- Main: Machine Learning (cs.LG)
- Secondary: Artificial Intelligence (cs.AI); Robotics (cs.RO)
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 paper introduces a novel approach to offline reinforcement learning (RL) that prioritizes safety constraints, ensuring zero violation of safety requirements. The existing methods focus on soft constraints, which may lead to potentially unsafe outcomes. To address this limitation, the authors develop FISOR (FeasIbility-guided Safe Offline RL), a framework that decouples three processes: safety constraint adherence, reward maximization, and offline policy learning. By leveraging reachability analysis from safe-control theory, FISOR translates the hard safety constraint into identifying the largest feasible region given the offline dataset. This allows for maximizing reward value within the feasible region while minimizing safety risks in the infeasible region. The authors demonstrate that FISOR achieves strong safety performance and stability on the DSRL benchmark for safe offline RL, outperforming baselines by guaranteeing zero violation of safety requirements in all tasks. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary FISOR is a new way to make sure that artificial intelligence systems behave safely when they’re not connected to the internet. Right now, most AI systems are only checked for how well they do on average, but this can sometimes lead to bad things happening even if it’s rare. FISOR makes sure that these systems never do anything bad by using a special kind of math called reachability analysis. This helps make sure that the AI system is always doing what we want it to do and not something else. The people who developed FISOR tested it on some standard tests and found that it worked really well, making sure that the AI systems were safe in all scenarios. |
Keywords
* Artificial intelligence * Reinforcement learning
