Research map / Reinforcement learning / Visual and representation RL
InternData-A1: Pioneering High-Fidelity Synthetic Data for Pre-training Generalist Policy
Yang Tian, Yuyin Yang, Yiman Xie, Zetao Cai, Xu Shi, Ning Gao, Hangxu Liu, Xuekun Jiang, Zherui Qiu, Feng Yuan, Yaping Li, Ping Wang and 4 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 |
|---|---|---|---|
| pretrain pi0 on InternData-A1 instead of official pi-dataset | pi0 · RoboTwin 2.0 49 tasks Easy | success rate: 55.0% -> 60.0% | Improved |
| pretrain on InternData-A1 instead of pi-dataset | pi0 · RoboTwin 2.0 49 tasks Hard | success rate: 20.0% -> 26.5% | Improved |
| pretrain on InternData-A1 instead of no pretraining | pi0 · RoboTwin 2.0 49 tasks | success rate: 23.5% -> 60.0% Easy; 2.5% -> 26.5% Hard | Improved |
| InternData-A1 vs pi-dataset on selected sim tasks | pi0 · Lift Pot Hard | success rate: 1.5% -> 2.5% | No clear effect |
| pretrain on InternData-A1 vs pi-dataset on real regular tasks | pi0 · Real-world 5 regular tasks | success rate: 6.2% margin | Improved |
| pretrain on InternData-A1 vs pi-dataset on dexterous tasks | pi0 · Real-world 4 dexterous tasks | success rate | No clear effect |
| pretrain on InternData-A1 instead of RoboCasa (sim) | pi0 · Real-world Sort Rubbish | success rate: 23.3% -> 90.0% | Improved |
| pretrain on InternData-A1 instead of Agibot World (real) | pi0 · 49 sim tasks Easy | success rate: 52.5% -> 60.0% | Improved |
| pretrain on InternData-A1 instead of Agibot World (real) | pi0 · Real-world Pass Bottle | success rate: 56.7% -> 60.0% | Improved |
| include base tasks in pretraining mix (ablation) | pi0 (InternData-A1) · RoboTwin 2.0 Easy | success rate: 52.5% -> 58.0% | Improved |
| include long-horizon tasks (ablation) | pi0 (InternData-A1) · RoboTwin 2.0 Hard | success rate: 19.0% -> 25.0% | Improved |
| include articulation data (ablation) | pi0 (InternData-A1) · RoboTwin 2.0 Hard | success rate: 19.5% -> 25.0% | Improved |
Closest papers by meaning
- GenSim: Generating Robotic Simulation Tasks via Large Language Models (ICLR 2024 · similarity 0.88)
- Scaling Proprioceptive-Visual Learning with Heterogeneous Pre-trained Transformers (NeurIPS 2024 · similarity 0.88)
- VLA-REPLICA: A Low-Cost, Reproducible Benchmark for Real-World Evaluation of Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- DexScale: Automating Data Scaling for Sim2Real Generalizable Robot Control (ICML 2025 · similarity 0.87)
- Beyond Data Scaling: Representation-Centric Pre-training for Vision-Language-Action Models (NeurIPS 2026 · similarity 0.87)
- Self-Improving Embodied Foundation Models (NeurIPS 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.
