Research map / Reinforcement learning / Visual and representation RL
Direct then Diffuse: Incremental Unsupervised Skill Discovery for State Covering and Goal Reaching
Pierre-Alexandre Kamienny, Jean Tarbouriech, sylvain lamprier, Alessandro Lazaric, Ludovic Denoyer
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 |
|---|---|---|---|
| Use direct-then-diffuse skills composed in a growing tree instead of DIAYN skills | UPSIDE · Bottleneck Maze | coverage | Improved |
| Use direct-then-diffuse skills composed in a growing tree instead of DIAYN skills | UPSIDE · Control environments | state coverage | Improved |
| Use UPSIDE's learned skill tree for fine-tuning sparse-reward goal reaching instead of DIAYN-20 or TD3 | UPSIDE · Ant, 48 unknown goals | success rate | Improved |
| Use UPSIDE's learned skill tree for fine-tuning sparse-reward goal reaching instead of other methods | UPSIDE · Bottleneck Maze, unknown goals | discounted cumulative reward | Improved |
| Remove tree composition by using flat UPSIDE with tree depth 1 (ablation) | UPSIDE · Bottleneck Maze | exploration | Hurt |
| Remove tree composition by using flat UPSIDE with tree depth 1 (ablation) | UPSIDE · Bottleneck Maze | coverage | Hurt |
| Keep the diffusing part in flat UPSIDE rather than using DIAYN (ablation) | UPSIDE · Coverage environments | coverage | Improved |
| Remove the direct-then-diffuse structure, scoring the entire directed-skill trajectory as in DIAYN-hier (ablation) | UPSIDE · Bottleneck Maze | performance | Hurt |
| Remove the direct-then-diffuse structure, scoring the entire directed-skill trajectory as in DIAYN-hier (ablation) | UPSIDE · Evaluated environments | performance | Hurt |
| Remove the discriminability-constrained policy-count selection while retaining hierarchical composition and diffusion, as in DIAYN-hier (ablation) | UPSIDE · Bottleneck Maze | sample efficiency of tree growth | Hurt |
| Use SAC rather than TD3 as the policy optimizer (ablation) | DIAYN · Bottleneck Maze | coverage: 21.0 (± 0.50) vs 17.67 (± 0.57) | No clear effect |
| Use SAC rather than TD3 as the policy optimizer (ablation) | DIAYN · U-maze | coverage: 17.5 (± 0.75) vs 14.67 (± 0.42) | No clear effect |
Closest papers by meaning
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.90)
- Exploration by Learning Diverse Skills through Successor State Representations (NeurIPS 2024 · similarity 0.90)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.90)
- Task Adaptation from Skills: Information Geometry, Disentanglement, and New Objectives for Unsupervised Reinforcement Learning (ICLR 2024 · similarity 0.90)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.89)
- Unifying Goal-Conditioned RL and Unsupervised Skill Learning via Control-Maximization (NeurIPS 2026 · 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.
