Research map / Reinforcement learning / Visual and representation RL
Behavior Contrastive Learning for 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 contrastive behavior objective (BeCL) instead of MI-based skill discovery | unsupervised skill discovery · URLB aggregate (12 tasks) | IQM normalized score: BeCL 75.38% vs CIC 75.18%, APS trailing by 38.2% | Helped |
| Use BeCL objective | unsupervised skill discovery · URLB aggregate | Optimality gap: 25.44% vs CIC 25.75% | Helped |
| Use BeCL objective instead of CIC (state-entropy) objective | Quadruped tasks · URLB Quadruped | normalized score: BeCL higher on all four Quadruped tasks (e.g., Stand 875 vs 700) | Helped |
| Use BeCL objective instead of CIC | Walker tasks · URLB Walker | normalized score: BeCL below CIC on Flip, Run, Stand, Walk | Mixed |
| Increase skill dimension in discrete skill space (BeCL) (ablation) | BeCL · tree-like maze and DMC Walker | skill coverage / adaptation | Helped |
| Lower temperature from 1 to 0.5 (kappa) (ablation) | BeCL · continuous 2D maze | maze coverage | Helped |
| Decrease temperature further from 0.1 to 0.01 (ablation) | BeCL · continuous 2D maze | skill distinguishability | Hurt |
| Choose skills in fine-tuning rather than random sampling (CIC skill selection) | CIC · URLB Quadruped fine-tuning | adaptation reward | No clear effect |
Closest papers by meaning
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.92)
- Constrained Ensemble Exploration for Unsupervised Skill Discovery (ICML 2024 · similarity 0.91)
- Efficient Skill Discovery via Regret-Aware Optimization (ICML 2025 · similarity 0.90)
- Disentangled Unsupervised Skill Discovery for Efficient Hierarchical Reinforcement Learning (NeurIPS 2024 · similarity 0.90)
- Dynamic Contrastive Skill Learning with State-Transition Based Skill Clustering and Dynamic Length Adjustment (ICLR 2025 · similarity 0.90)
- Exploration by Learning Diverse Skills through Successor State Representations (NeurIPS 2024 · 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.