← Leaderboard

RDT2

VLA
Tsinghua University · Sep 2025

Successor to RDT-1B trained on more than 10,000 hours of handheld UMI gripper data for zero-shot cross-embodiment transfer, released as RDT2-VQ and RDT2-FM. Results are self-reported on the authors’ own real-robot tasks (77% on cloth folding after fine-tuning vs 36% for π0.5); weights came out in September 2025, the paper in February 2026.

Type
VLA
Parameters
8.3B
Backbone
Qwen2.5-VL-7B (RDT2-VQ); RDT2-FM adds a flow-matching action expert
Weights
Open
License
Apache-2.0
Commercial use
Allowed
Released
Sep 2025
Robots
Bimanual UR5e, Bimanual Franka FR3, UMI
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

Not reported on 13 benchmarks
Wrong or missing number? Report an error