Research map / Reinforcement learning / Visual and representation RL
Controllability-Aware Unsupervised Skill Discovery
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 a jointly trained controllability-aware transition distance instead of Euclidean distance or mutual information | CSD · FetchPush, FetchSlide, and FetchPickAndPlace | object manipulation skill discovery | Helped |
| Use a jointly trained controllability-aware transition distance instead of Euclidean distance or mutual information | CSD · Kitchen, 16 discrete skills | downstream tasks with at least one successful skill: 10 out of 13; 2 for LSD, 4 for DIAYN, 0 for DADS | Helped |
| Use a jointly trained controllability-aware transition distance | CSD · Ant and HalfCheetah | state coverage | Helped |
| Use a jointly trained controllability-aware transition distance | CSD · AntGoal, AntMultiGoals, HalfCheetahGoal, and HalfCheetahHurdle | downstream task performance | Helped |
| Use a jointly trained controllability-aware transition distance | CSD · Fetch manipulation downstream tasks | goal-reaching performance | Helped |
| Normalize state dimensions using precomputed standard deviations from random trajectories (ablation) | LSD · Fetch downstream tasks | object manipulation skill discovery | No clear effect |
| Normalize state dimensions using a moving average of state-difference standard deviations (ablation) | LSD · Fetch downstream tasks | object manipulation skill discovery | No clear effect |
| Replace spectral normalization with dual gradient descent, without adding a learned distance function (ablation) | LSD · Fetch downstream tasks | object manipulation skill discovery | No clear effect |
| Add the learned controllability-aware distance function (ablation) | LSD with dual gradient descent · Fetch downstream tasks | object manipulation skill discovery | Helped |
| Use CSD rather than LSD with the oracle state prior | CSD and LSD · FetchPickAndPlace, oracle prior | runs discovering object-pickup skills: all eight runs; only three out of eight runs | Helped |
Closest papers by meaning
- Periodic Skill Discovery (NeurIPS 2025 · similarity 0.88)
- SkiLD: Unsupervised Skill Discovery Guided by Factor Interactions (NeurIPS 2024 · similarity 0.88)
- Disentangled Unsupervised Skill Discovery for Efficient Hierarchical Reinforcement Learning (NeurIPS 2024 · similarity 0.88)
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.88)
- Language Guided Skill Discovery (ICLR 2025 · similarity 0.87)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · 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.