Research map / Reinforcement learning / MDPs and planning
A Simple Reward-free Approach to Constrained Reinforcement Learning
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Use the reward-free meta-algorithm to solve approachability with a tabular reward-free solver | Tabular VMDPs · Approachability | sample complexity: $\tilde{\mathcal{O}}(\min\{d, S\} H^4SA/\epsilon^2)$ versus $\tilde{\mathcal{O}}(\min\{d, S\} H^3SA/\epsilon^2)$ | Mixed |
| Use the reward-free meta-algorithm to solve CMDP with general convex constraints | Tabular VMDPs · CMDP | sample complexity: $\tilde{\mathcal{O}}(\min\{d, S\} H^4SA/\epsilon^2)$ versus $\tilde{\mathcal{O}}(\min\{d, S\} H^3SA/\epsilon^2)$ | Mixed |
| Use reward-free exploration with the approachability reduction | Linear VMDPs · Approachability and CMDP | sample complexity: $\tilde{\mathcal{O}}(d_{\text{lin}}^3 H^6/\epsilon^2)$ | Helped |
| Extend the reward-free framework to two-player vector-valued Markov games | Tabular VMGs · Blackwell approachability | regret rate: $\alpha(T)= \mathcal{O}(\epsilon/2 + \sqrt{H^2\iota/T})$ | Mixed |
| Use reward-free exploration rather than a regret-based approach | Tabular constrained RL · CMDP | regret guarantees | Hurt |
Closest papers by meaning
- Model-based RL as a Minimalist Approach to Horizon-Free and Second-Order Bounds (ICLR 2025 · similarity 0.91)
- Reward Penalties on Augmented States for Solving Richly Constrained RL Effectively (AAAI 2024 · similarity 0.90)
- Provably Efficient RL under Episode-Wise Safety in Constrained MDPs with Linear Function Approximation (NeurIPS 2025 · similarity 0.90)
- Constrained Reinforcement Learning Under Model Mismatch (ICML 2024 · similarity 0.90)
- A Primal-Dual Algorithm for Offline Constrained Reinforcement Learning with Linear MDPs (ICML 2024 · similarity 0.89)
- To the Max: Reinventing Reward in Reinforcement Learning (ICML 2024 · similarity 0.89)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.