Research map / Reinforcement learning / Visual and representation RL
URLB: Unsupervised Reinforcement Learning Benchmark
Michael Laskin, Denis Yarats, Hao Liu, Kimin Lee, Albert Zhan, Kevin Lu, Catherine Cang, Lerrel Pinto, Pieter Abbeel
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| unsupervised pre-training up to 2M steps then 100k fine-tune vs best methods | all eight baselines · URLB (12 tasks) | normalized return: 40-70% normalized return vs ~100% needed | Mixed |
| pre-train with intrinsic rewards instead of random initialization (states) (ablation) | DDPG · URLB states | extrinsic return | No clear effect |
| pre-train with intrinsic rewards instead of random initialization (pixels) (ablation) | DrQ-v2 · URLB pixels | extrinsic return: Walker Stand 212 -> 552 (Disagreement, 2M) | Improved |
| switch observations from states to pixels | exploration algorithms · URLB | normalized return: lose 20-50% | Hurt |
| competence-based vs knowledge/data-based intrinsic rewards | unsupervised RL · URLB | aggregate normalized return | Hurt |
| increase pre-training steps from 100k to 2M (ablation) | APS · Walker Flip, states | extrinsic return: 638 -> 461 | Hurt |
| increase pre-training steps from 100k to 2M (ablation) | Disagreement · Quadruped Walk, states | extrinsic return: 265 -> 721 | Improved |
| increase pre-training steps (ablation) | unsupervised RL algorithms · URLB, 18 experiments | fine-tuning performance: 9 of 18 do not improve or degrade | Mixed |
Closest papers by meaning
- How Far Can Unsupervised RLVR Scale LLM Training? (ICLR 2026 · similarity 0.88)
- Towards General-Purpose Model-Free Reinforcement Learning (ICLR 2025 · similarity 0.87)
- PEAC: Unsupervised Pre-training for Cross-Embodiment Reinforcement Learning (NeurIPS 2024 · similarity 0.87)
- Foundation Policies with Hilbert Representations (ICML 2024 · similarity 0.87)
- Is Inverse Reinforcement Learning Harder than Standard Reinforcement Learning? A Theoretical Perspective (ICML 2024 · similarity 0.86)
- Dual RL: Unification and New Methods for Reinforcement and Imitation Learning (ICLR 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.
