Research map / Video / Action recognition
Open-World Skill Discovery from Unsegmented Demonstration Videos
Jingwen Deng, Zihao Wang, Shaofei Cai, Anji Liu, Yitao Liang
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 |
|---|---|---|---|
| retrain GROOT on SBD-segmented data instead of random splitting | GROOT · Minecraft skill benchmark | relative average performance: +63.7% | Improved |
| retrain STEVE-1 on SBD-segmented data | STEVE-1 · Minecraft skill benchmark | relative average performance: +52.1% | Improved |
| use SBD-trained controller in OmniJARVIS | OmniJARVIS · long-horizon programmatic tasks | relative average performance: +11.3% | Improved |
| use SBD-trained controller in JARVIS-1 | JARVIS-1 · long-horizon programmatic tasks | relative average performance: +20.8% | Improved |
| add external event information to random splitting (ablation) | GROOT · Minecraft skill benchmark | normalized average score: 63.3 -> 85.8 | Improved |
| use prediction-loss boundaries instead of external info only (ablation) | GROOT SBD · Minecraft skill benchmark | normalized average score: 85.8 -> 90.2 | Improved |
| combine loss and external information (ablation) | GROOT SBD · Minecraft skill benchmark | normalized average score: 90.2 -> 91.7 | Improved |
| vary GAP from 17.5 to 18.5 (ablation) | GROOT SBD loss only · Minecraft skill benchmark | normalized average score: 88.4 / 90.2 / 88.5 | No clear effect |
| random splitting that matches SBD segment length distribution (ablation) | GROOT · Minecraft skill benchmark | normalized average score: 63.3 -> 79.9 | Improved |
| use SBD instead of length-matched random splitting (ablation) | GROOT · Minecraft skill benchmark | normalized average score: 79.9 -> 91.7 | Improved |
| combine loss and info for use furnace task (ablation) | GROOT SBD · Minecraft use furnace | success rate: 44.0% -> 30.0% | Hurt |
| SBD in combat scenes | GROOT SBD · Minecraft combat spiders | success rate: 29% -> 20% | Hurt |
Closest papers by meaning
- GROOT: Learning to Follow Instructions by Watching Gameplay Videos (ICLR 2024 · similarity 0.87)
- Data Augmentation for Instruction Following Policies via Trajectory Segmentation (AAAI 2025 · similarity 0.86)
- SLVMBench: Skill Learning from Video Memory (NeurIPS 2026 · similarity 0.86)
- Do's and Don'ts: Learning Desirable Skills with Instruction Videos (NeurIPS 2024 · similarity 0.85)
- Language-guided Skill Learning with Temporal Variational Inference (ICML 2024 · similarity 0.85)
- Periodic Skill Discovery (NeurIPS 2025 · similarity 0.84)
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.
