Research map / Reinforcement learning / Imitation learning
Making Learner Weakness Actionable for Learning from Demonstration with Novice Teachers
Yuqing Zhu, Matthew Howard
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 |
|---|---|---|---|
| add CLASP-Teach demonstration guidance and difficulty-aware reweighting (CLASP-Full) | diffusion policy · Robust success, 4 tasks at B=20 | success rate under shift: Avoiding 0.62 -> 0.88; Pushing 0.34 -> 0.72; P&P 0.44 -> 0.78; Alignment 0.16 -> 0.54 (vs no guidance) | Improved |
| add difficulty-aware reweighting (CLASP-Train) on top of CLASP-Teach (ablation) | CLASP · Alignment, B=20 | success rate: 0.56 -> 0.66 | Improved |
| add difficulty-aware reweighting (CLASP-Train) on top of CLASP-Teach (ablation) | CLASP · Avoiding, B=20 | success rate: 0.92 -> 0.94 | No clear effect |
| add difficulty-aware reweighting (CLASP-Train) on top of CLASP-Teach (ablation) | CLASP · Alignment, robust success under shift, B=20 | success rate: 0.42 -> 0.54 | Improved |
| use CLASP guidance instead of uncertainty-gated querying (ASkDAgger) | diffusion policy · Robust success, Alignment B=20 | success rate under shift: 0.22 -> 0.54 | Improved |
| offline curation (quality or cluster weighting) without new guided demonstrations | no-guidance dataset · Robust success, Avoiding B=20 | success rate under shift: 0.62 -> 0.68 (quality), 0.70 (cluster); Alignment 0.16 -> 0.20, 0.18 | Mixed |
| easy-to-hard curriculum for demonstration collection | diffusion policy · Robust success, Pushing B=20 | success rate under shift: 0.34 -> 0.58 | Improved |
| CLASP guidance on real robot | diffusion policy · Real-robot Avoiding, B=20 | success rate: 0.76 -> 0.92 | Improved |
| CLASP guidance on real robot | diffusion policy · Real-robot Stirring, B=20 | success rate: 0.68 -> 0.76 | Improved |
| CLASP guidance at very small budget on real robot | diffusion policy · Real-robot Avoiding, B=4 | success rate: 0.38 -> 0.32 | Hurt |
| CLASP guidance on completion time for easy task | diffusion policy · Avoiding simulation, B=20 | normalised completion time: 0.34 -> 0.32 | No clear effect |
| add CLASP guidance for behavioural diversity | diffusion policy · Avoiding and Pushing strategy balance | balance score r | Improved |
Closest papers by meaning
- When a Robot is More Capable than a Human: Learning from Constrained Demonstrators (ICLR 2026 · similarity 0.88)
- Learning Shared Safety Constraints from Multi-task Demonstrations (NeurIPS 2023 · similarity 0.85)
- Student-Informed Teacher Training (ICLR 2025 · similarity 0.85)
- Learning from Demonstrations via Capability-Aware Goal Sampling (NeurIPS 2025 · similarity 0.84)
- A Smooth Sea Never Made a Skilled SAILOR: Robust Imitation via Learning to Search (NeurIPS 2025 · similarity 0.84)
- Uncertainty-Based Smooth Policy Regularisation for Reinforcement Learning with Few Demonstrations (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.
