Research map / Reinforcement learning / Imitation learning
Twin-DAgger: Synergizing Digital Twins and Human Corrections for Efficient Robot Manipulation
Jiahang Li, Zhirui Zhang, Kairan Ding, Fan Fei, Yunkai Tang, Jiaming Liu, Xiao He, Ziyu Chen, Gaohao Zhou, Jieji Ren, Yandong Guo, Shanghang Zhang and 1 more
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 difficulty-based stratified sampling (ablation) | Twin-DAgger (ACT) · Pick-and-place (real robot) | success rate: 46.7% -> 55.0% | Improved |
| add difficulty-based stratified sampling (ablation) | Twin-DAgger (pi0) · Pick-and-place (real robot) | success rate: 68.3% -> 85.0% | Improved |
| add intervention-focused trajectory warping (ablation) | Twin-DAgger (ACT) · Pick-and-place (real robot) | success rate: 48.3% -> 55.0% | Improved |
| add intervention-focused trajectory warping (ablation) | Twin-DAgger (pi0) · Pick-and-place (real robot) | success rate: 76.7% -> 85.0% | Improved |
| scale Real2Render2Real from 100 to 500 augmented samples | pi0 · Pick-and-place (real robot) | success rate: 55.0% -> 43.3% | Hurt |
| scale Real2Render2Real from 100 to 500 augmented samples | ACT · Pick-and-place (real robot) | success rate: 33.0% -> 45.0% | Improved |
| raise concentrating factor gamma from 5 to 10 (ablation) | Twin-DAgger (ACT) · Pick-and-place (real robot) | success rate: 55.0% -> 38.3% | Hurt |
| raise temporal perturbation scale from 1 to 5 (ablation) | Twin-DAgger (ACT) · Pick-and-place (real robot) | success rate: 55.0% -> 40% | Hurt |
| lower temporal perturbation scale from 1 to 0.5 (ablation) | Twin-DAgger (ACT) · Pick-and-place (real robot) | success rate: 55.0% -> 51.7% | Hurt |
| add more teleoperation data (30 extra) | ACT · Pick-and-place (real robot) | success rate: 30.0% -> 40.0% | No clear effect |
| add DT augmentation with randomized friction (ablation) | ACT · Varied friction (real) | success rate: 15.0% -> 20.0% | Improved |
| Twin-DAgger vs conventional DAgger on Peg-in-Hole | ACT · Peg-in-Hole | success rate: 46.7% -> 43.3% | Hurt |
Closest papers by meaning
- DexFlyWheel: A Scalable and Self-improving Data Generation Framework for Dexterous Manipulation (NeurIPS 2025 · similarity 0.90)
- Compliant Residual DAgger: Improving Real-World Contact-Rich Manipulation with Human Corrections (NeurIPS 2025 · similarity 0.90)
- Data Scaling Laws in Imitation Learning for Robotic Manipulation (ICLR 2025 · similarity 0.88)
- AHA: A Vision-Language-Model for Detecting and Reasoning Over Failures in Robotic Manipulation (ICLR 2025 · similarity 0.88)
- Generalizable Domain Adaptation for Sim-and-Real Policy Co-Training (NeurIPS 2025 · similarity 0.87)
- Dream to Manipulate: Compositional World Models Empowering Robot Imitation Learning with Imagination (ICLR 2025 · similarity 0.87)
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.
