Research map / Probabilistic and scientific modeling / Molecules and proteins
COLLIE: Guiding Skill Discovery in Semantically Coherent Latent Space
Yao Luan, Ni Mu, Hanfei Ge, Yiqin Yang, Bo XU, Qing-Shan Jia
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 |
|---|---|---|---|
| build guidance in a semantically coherent skill latent space instead of Euclidean raw-state distance (ablation) | COLLIE · Ant North | safe state coverage: 657.60 -> 1333.20 | Improved |
| build guidance in a semantically coherent skill latent space instead of Euclidean raw-state distance (ablation) | COLLIE · HalfCheetah Not-Flip | safe state coverage: 104.20 -> 215.60 | Improved |
| increase number of human feedback labels from 10 to 40 (ablation) | COLLIE · Ant North | safe state coverage: 801.40 -> 1333.20 | Improved |
| increase number of human feedback labels from 10 to 40 (ablation) | COLLIE · Ant Hole | safe state coverage: 766.20 -> 1149.20 | Improved |
| add labeling noise near safety boundaries (Rerror 0 to 1) (ablation) | COLLIE · Ant North | safe state coverage: 1333.20 -> 1084.20 | Hurt |
| add labeling noise near safety boundaries (Rerror 0 to 1) (ablation) | COLLIE · Ant Range | safe state coverage: 362.20 -> 360.80 | No clear effect |
| use training-free guidance signal in latent space instead of trained instruction network (DoDont*) | guided skill discovery · Ant Range | safe state coverage: -427.60 -> 362.20 | Improved |
| add COLLIE guidance to METRA | METRA · Ant North | safe state coverage: -1425.80 -> 1333.20 | Improved |
| add COLLIE guidance to METRA | METRA · Ant Hole | safe state coverage: 1179.00 -> 1149.20 | No clear effect |
| use COLLIE skills for zero-shot downstream Ant Hole task | METRA · Ant Hole zero-shot (avg) | task return: 224.07 -> 650.67 | Improved |
| use COLLIE skills with a hierarchical controller | METRA · HalfCheetah downstream | performance: 21.58 -> 45.26 | Improved |
Closest papers by meaning
- Language Guided Skill Discovery (ICLR 2025 · similarity 0.89)
- Skill Disentanglement in Reproducing Kernel Hilbert Space (AAAI 2025 · similarity 0.88)
- SkiLD: Unsupervised Skill Discovery Guided by Factor Interactions (NeurIPS 2024 · similarity 0.86)
- Language-guided Skill Learning with Temporal Variational Inference (ICML 2024 · similarity 0.86)
- Do's and Don'ts: Learning Desirable Skills with Instruction Videos (NeurIPS 2024 · similarity 0.86)
- Disentangled Unsupervised Skill Discovery for Efficient Hierarchical Reinforcement Learning (NeurIPS 2024 · similarity 0.86)
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.
