Research map / Reinforcement learning / Policy gradient and actor-critic
Mirror Descent Policy Optimization
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 on-policy MDPO instead of PPO | On-policy policy optimization · Six MuJoCo continuous-control tasks | average cumulative return | Helped |
| Use on-policy MDPO instead of TRPO | On-policy policy optimization · Six MuJoCo continuous-control tasks | average cumulative return | Mixed |
| Use off-policy MDPO-KL instead of SAC | Off-policy policy optimization · Six MuJoCo continuous-control tasks | average cumulative return | No clear effect |
| Use Tsallis rather than KL Bregman divergence | Off-policy MDPO · Loaded MuJoCo tasks | average cumulative return | Mixed |
| Use multiple gradient steps per on-policy MD iteration rather than one (ablation) | On-policy MDPO · Walker2d-v2 and other evaluated tasks | performance | Helped |
| Use multiple gradient steps per TRPO iteration rather than a single-step update (ablation) | TRPO · Evaluated tasks | performance | No clear effect |
| Increase the number of gradient steps in off-policy MDPO (ablation) | Off-policy MDPO · Hopper-v2 and evaluated tasks | score | Helped |
| Use the same q for the MDP regularizer and Bregman divergence (ablation) | Off-policy MDPO-Tsallis · Walker2d-v2, Ant-v2, and Humanoid-v2 | average cumulative return | Helped |
| Tune q in SAC-Tsallis (ablation) | SAC · Evaluated MuJoCo tasks | performance | No clear effect |
Closest papers by meaning
- Reflective Policy Optimization (ICML 2024 · similarity 0.91)
- Embedding Safety into RL: A New Take on Trust Region Methods (ICML 2025 · similarity 0.90)
- Simple Policy Optimization (ICML 2025 · similarity 0.90)
- Policy Mirror Descent with Lookahead (NeurIPS 2024 · similarity 0.89)
- ACPO: A Policy Optimization Algorithm for Average MDPs with Constraints (ICML 2024 · similarity 0.89)
- Learning mirror maps in policy mirror descent (ICLR 2025 · 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.