Research map / Reinforcement learning / Visual and representation RL
Explore, Discover and Learn: Unsupervised Discovery of State-Covering Skills
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 EDL with a fixed state distribution and variational modeling instead of policy-induced state distributions | Information-theoretic skill discovery · 2D maze with bottleneck states | state-space coverage | Helped |
| Replace oracle exploration with SMM exploration (ablation) | EDL · Simple maze | state-space coverage | No clear effect |
| Use SMM exploration in a maze with bottleneck states (ablation) | EDL · Maze with bottleneck states | exploration coverage | Hurt |
| Use oracle exploration instead of SMM in the bottleneck maze (ablation) | EDL · Maze with bottleneck states | state-space coverage | Helped |
| Add Sibling Rivalry during skill learning (ablation) | EDL · Maze with bottleneck states | successful skill learning | Helped |
| Change the initial-state distribution between E_center and E_left (ablation) | Forward and reverse baselines · Corridor mazes | similarity of discovered skills | Hurt |
| Change the learning environment's initial-state distribution between E_center and E_left (ablation) | EDL with SMM exploration · Corridor mazes | similarity of discovered skills | No clear effect |
| Set the state prior to focus on the right side of the maze (ablation) | EDL · Tree-shaped maze | coverage of the region of interest | Helped |
| Use a Gaussian assumption for p(s|z) in the state density model (ablation) | EDL · Maze skill learning | reward landscape quality | Hurt |
| Interpolate between learned latent codes | EDL skill policies · Maze environments | generalization to unseen skills | Helped |
| Use a fixed state distribution rather than a policy-induced state distribution | Information-theoretic skill discovery · State-space coverage | reward for visiting previously unseen states: r_max = log N; r_new = 0 | Helped |
Closest papers by meaning
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.91)
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.91)
- When Do Skills Help Reinforcement Learning? A Theoretical Analysis of Temporal Abstractions (ICML 2024 · similarity 0.90)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.89)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.89)
- DISCOVER: Automated Curricula for Sparse-Reward Reinforcement Learning (NeurIPS 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.