Research map / Vision-language and multimodal / Robot action and navigation
RDT2: Exploring the Scaling Limit of UMI Data Towards Zero-Shot Cross-Embodiment Generalization
Songming Liu, Bangguo Li, Kai Ma, Lingxuan Wu, Hengkai Tan, Xiao Ouyang, Hang Su, Jun Zhu
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 AR (RVQ cross-entropy) pretraining stage before diffusion (ablation) | RDT2 · Training loss | convergence speed and final loss | Improved |
| replace FAST tokenizer with RVQ tokenizer (ablation) | RDT2 action discretization · Discretization error vs tokens | tokens needed at equal error: saves up to about two-thirds of the tokens | Improved |
| use uniform binning instead of RVQ (ablation) | RDT2 action discretization · Discretization error vs tokens | token efficiency: lowest error but far more tokens | Hurt |
| add one-step diffusion distillation (Stage 3) (ablation) | RDT2-FM · Inference frequency | Hz: 16.0 Hz -> 23.0 Hz | Improved |
| add flow-matching action expert (Stage 2) over RDT2-VQ (ablation) | RDT2-VQ · Inference frequency | Hz: 2.7 Hz -> 16.0 Hz | Improved |
| flow-matching expert vs RVQ autoregressive head (ablation) | RDT2 · Zero-shot tasks (unseen embodiment) | success rate | No clear effect |
| scale model size and data jointly (ablation) | RDT2-VQ · Training loss (one epoch) | training loss | Improved |
| fine-tune RDT2 instead of pi0.5 | pi0.5 · Cloth folding | success rate (%): 36 -> 77 | Improved |
| fine-tune RDT2 instead of pi0.5 | pi0.5 · Unzipping | success rate (%): 13 -> 45 | Improved |
| fine-tune RDT2 instead of pi0.5 | pi0.5 · Button pressing | reaction time gap (ms): +323 -> +97 | Improved |
Closest papers by meaning
- Exploring the Limits of Vision-Language-Action Manipulation in Cross-task Generalization (NeurIPS 2025 · similarity 0.92)
- Sim2Real VLA: Zero-Shot Generalization of Synthesized Skills to Realistic Manipulation (ICLR 2026 · similarity 0.91)
- NS-VLA: Towards Neuro-Symbolic Vision-Language-Action Models (NeurIPS 2026 · similarity 0.89)
- RealVLG-R1: A Large-Scale Real-World Visual-Language Grounding Benchmark for Robotic Perception and Manipulation (CVPR 2026 · similarity 0.89)
- OmniManip: Towards General Robotic Manipulation via Object-Centric Interaction Primitives as Spatial Constraints (CVPR 2025 · similarity 0.89)
- Vision-Language Foundation Models as Effective Robot Imitators (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.
