Research map / Reinforcement learning / Visual and representation RL
Benchmarking Deep Reinforcement Learning for Continuous Control
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 TRPO instead of REINFORCE | policy gradient · Hopper | average return: 714.0 -> 1183.3 | Helped |
| use TNPG instead of REINFORCE | policy gradient · 2D Walker | average return: 506.5 -> 1382.6 | Helped |
| use TRPO line search instead of TNPG | natural gradient · Swimmer | average return: 96.0 -> 96.0 | No clear effect |
| use TRPO larger step size delta_KL 0.1 vs TNPG best 0.05 | trust-region update · Swimmer | learning speed: 0.05 -> 0.1 | Helped |
| use DDPG instead of TNPG | off-policy actor-critic · Half-Cheetah | average return: 1729.5 -> 2148.6 | Helped |
| use DDPG instead of TRPO | off-policy actor-critic · Hopper | average return: 1183.3 -> 267.1 | Hurt |
| rescale rewards by 0.1 for DDPG | DDPG · all tasks | stability: factor of 0.1 | Helped |
| use CEM instead of CMA-ES | gradient-free search · basic tasks | average return: 2440.4 -> 4815.4 | Helped |
| use CMA-ES on high-dimensional observations | CMA-ES · Full Humanoid | average return | Hurt |
| use recurrent policy for CEM and CMA-ES in partially observable tasks | gradient-free methods · Cart-Pole Balancing (LS) | average return: 4815.4 -> 227.0 | Hurt |
| use recurrent policies instead of feed-forward | policy · partially observable tasks | solution quality | Mixed |
| use TNPG instead of REINFORCE with recurrent policy | policy gradient · Cart-Pole Balancing (SI) | average return: 431.7 -> 980.5 | Helped |
Closest papers by meaning
- Deep Reinforcement Learning at the Edge of the Statistical Precipice (NeurIPS 2021 · similarity 0.89)
- Mastering Visual Continuous Control: Improved Data-Augmented Reinforcement Learning (ICLR 2022 · similarity 0.88)
- Position: Benchmarking is Limited in Reinforcement Learning Research (ICML 2024 · similarity 0.87)
- Deep Reinforcement Learning for Robotics: A Survey of Real-World Successes (AAAI 2025 · similarity 0.87)
- XLand-100B: A Large-Scale Multi-Task Dataset for In-Context Reinforcement Learning (ICLR 2025 · similarity 0.87)
- Deep Policy Gradient Methods Without Batch Updates, Target Networks, or Replay Buffers (NeurIPS 2024 · similarity 0.86)
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.