Summary of Constrainedzero: Chance-constrained Pomdp Planning Using Learned Probabilistic Failure Surrogates and Adaptive Safety Constraints, by Robert J. Moss et al.
ConstrainedZero: Chance-Constrained POMDP Planning using Learned Probabilistic Failure Surrogates and Adaptive Safety Constraints
by Robert J. Moss, Arec Jamgochian, Johannes Fischer, Anthony Corso, Mykel J. Kochenderfer
First submitted to arxiv on: 1 May 2024
Categories
- Main: Artificial Intelligence (cs.AI)
- Secondary: None
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 ConstrainedZero policy iteration algorithm solves chance-constrained partially observable Markov decision processes (CC-POMDPs) in belief space, learning neural network approximations of the optimal value and policy. This is achieved through an additional network head that estimates the failure probability given a belief. The algorithm is combined with online Monte Carlo tree search (MCTS) for safe action selection, and adaptive conformal inference updates the failure threshold during planning to avoid overemphasizing search based on failure estimates. The approach is tested on several benchmarks, including a safety-critical POMDP benchmark, an aircraft collision avoidance system, and the sustainability problem of safe CO2 storage, achieving a target level of safety without optimizing the balance between rewards and costs. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary In this paper, researchers developed a new way to plan safely in uncertain situations. They created an algorithm called ConstrainedZero that helps agents make good decisions while keeping them safe. The algorithm uses neural networks to learn what actions are best and how likely they might fail. It also updates its failure threshold during planning to avoid making too many mistakes. The researchers tested their approach on several real-world problems, including avoiding plane crashes and storing CO2 safely. They found that it can achieve a target level of safety without sacrificing rewards. |
Keywords
» Artificial intelligence » Inference » Neural network » Probability
