Safe Reinforcement Learning with Preference-based Constraint Inference

Forum topic · 小凯 · 2026-03-27

Summary

This paper, authored by Chenglin Li, Guangchun Ruan, and Hua Geng and posted to arXiv (2603.23565), addresses a key challenge in safe reinforcement learning: how to handle safety constraints in safety-critical decision making when those constraints are complex, subjective, or difficult to specify explicitly. The authors note that existing constraint inference methods rely on restrictive assumptions or require extensive expert demonstrations, which is impractical in many real-world applications. The proposed approach instead infers constraints from preference-based feedback, offering a more flexible framework for safe RL in realistic settings. The work falls within the machine learning domain and was collected on 2026-03-27 from a Chinese tech forum discussion of the arXiv paper.

Overview

This paper was shared on zhichai.net and discusses safe reinforcement learning (RL) with preference-based constraint inference.

Paper details

Field: Machine Learning (ML)
Authors: Chenglin Li, Guangchun Ruan, Hua Geng
Published: 2026-03-26
arXiv: 2603.23565

Key points

Safe reinforcement learning is a standard paradigm for safety-critical decision making.
Real-world safety constraints can be complex, subjective, and even hard to explicitly specify.
Existing works on constraint inference rely on restrictive assumptions or extensive expert demonstrations, which is not realistic in many real-world applications.

Abstract (original)

Safe reinforcement learning (RL) is a standard paradigm for safety-critical decision making. However, real-world safety constraints can be complex, subjective, and even hard to explicitly specify. Existing works on constraint inference rely on restrictive assumptions or extensive expert demonstrations, which is not realistic in many real-world applications.

---

*Auto-collected on 2026-03-27.*