A Comprehensive Survey on Safe Reinforcement Learning

Provides a canonical overview or reference point for the relevant DoOperator research area.

A classic survey of safe reinforcement learning, including risk-sensitive criteria, constrained exploration, safety during learning, and external guidance.

Use when orienting a new paper, blog post, benchmark, or research plan in this area.

Do not cite the overview as evidence that a specific method works in a specific deployment setting without checking the underlying primary paper.