Research map / Reinforcement learning / Visual and representation RL
Choreographer: Learning and Adapting Skills in Imagination
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 |
|---|---|---|---|
| Add a meta-controller that evaluates and selects among skills (ablation) | Choreographer · URLB Walker and Quadruped tasks | fine-tuning performance | No clear effect |
| Add a meta-controller that considers all skills (ablation) | Choreographer · URLB Jaco tasks | fine-tuning performance | Helped |
| Fine-tune the skill policies during downstream-task adaptation (ablation) | Choreographer · Pixel-based URLB | fine-tuning performance | Helped |
| Use code resampling to keep skill codes active (ablation) | Choreographer · Pixel-based URLB | fine-tuning performance | Helped |
| Use code resampling (ablation) | Choreographer · Skill codebook training | reconstruction loss | Helped |
| Use Choreographer's latent-state skill objective, code resampling, and meta-controller instead of EDL's objective without those components (ablation) | Choreographer · URLB state-based and pixel-based | performance | Helped |
| Use skills with a lower sampling frequency (ablation) | Choreographer meta-controller · Downstream-task adaptation | adaptation performance | No clear effect |
| Use a Gaussian continuous skill representation instead of a discrete codebook (ablation) | Choreographer · Downstream-task adaptation | adaptation performance | Hurt |
| Use a standard discrete VAE instead of a VQ-VAE (ablation) | Choreographer skill discovery · Skill representation training | training stability | Hurt |
| Increase skill count beyond the tested effective range under the same compute budget (ablation) | Choreographer · Skill learning and adaptation | skill usefulness and fine-tuning | Hurt |
Closest papers by meaning
- Behavioral Exploration: Learning to Explore via In-Context Adaptation (ICML 2025 · similarity 0.88)
- Policy Compatible Skill Incremental Learning via Lazy Learning Interface (NeurIPS 2025 · similarity 0.87)
- RoboGen: Towards Unleashing Infinite Data for Automated Robot Learning via Generative Simulation (ICML 2024 · similarity 0.87)
- Skill Expansion and Composition in Parameter Space (ICLR 2025 · similarity 0.87)
- GROVE: A Generalized Reward for Learning Open-Vocabulary Physical Skill (CVPR 2025 · similarity 0.87)
- A Single Goal is All You Need: Skills and Exploration Emerge from Contrastive RL without Rewards, Demonstrations, or Subgoals (ICLR 2025 · similarity 0.87)
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.