Research map / Reinforcement learning / Imitation learning
Data Quality in Imitation Learning
Suneel Belkhale, Yuchen Cui, Dorsa Sadigh
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 |
|---|---|---|---|
| more system noise in expert data (0.05 to 0.2) (ablation) | BC · Square 200 episodes | success rate: 55.7(5.3) -> 82.0(3.4) | Improved |
| too much system noise in expert data (0.2 to 0.4) (ablation) | BC · Square 200 episodes | success rate: 82.0(3.4) -> 44.0(5.3) | Hurt |
| more system noise in expert data (0.05 to 0.2) (ablation) | BC · Square 50 episodes | success rate: 40.0(1.2) -> 69.7(0.7) | Improved |
| system noise 0.04 vs 0.01 in data (ablation) | BC · PMObstacle 1000 episodes | success rate: 97.7(1.5) -> 96.7(0.7) | No clear effect |
| add policy noise to the expert instead of system noise (ablation) | BC · Square 200 episodes | success rate: 82.0(3.4) -> 73.0(3.9) | Mixed |
| add policy noise with only 50 episodes (ablation) | BC · Square 50 episodes | success rate: 69.7(0.7) -> 57.7(3.2) | Hurt |
| longer horizon and higher action variance of human data | BC · Square | success rate: 58 -> 2 | Hurt |
| worse human operators on Can | BC · robomimic Can | success rate: 96 -> 22 | Hurt |
| state diversity as predictor of performance | BC · robomimic Square/Can | success rate | No clear effect |
| policy noise in expert data (PMObstacle) instead of system noise (ablation) | BC · PMObstacle 1000 episodes | success rate: 98.0(0.5) -> 92.3(2.0) | Hurt |
Closest papers by meaning
- How to Leverage Diverse Demonstrations in Offline Imitation Learning (ICML 2024 · similarity 0.91)
- Data Scaling Laws in Imitation Learning for Robotic Manipulation (ICLR 2025 · similarity 0.90)
- What Matters in Learning from Large-Scale Datasets for Robot Manipulation (ICLR 2025 · similarity 0.89)
- Is Behavior Cloning All You Need? Understanding Horizon in Imitation Learning (NeurIPS 2024 · similarity 0.89)
- Mitigating Covariate Shift in Behavioral Cloning via Robust Stationary Distribution Correction (NeurIPS 2024 · similarity 0.89)
- CCIL: Continuity-Based Data Augmentation for Corrective Imitation Learning (ICLR 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.
