Tsinghua

RDT (per-task)↗

RDT (per-task) is a VLA from Tsinghua, released October 10, 2024. It has no overall rank on the Robot Intelligence Index, which needs both a simulation and a real-robot result. Its best result is #16 of 22 on RoboChallenge. 1.2B diffusion-transformer foundation model for bimanual manipulation, predicting 64-step action chunks. (Fine-tuned separately per task.)

Release date: Oct 10, 2024

Developer
Tsinghua
Type
VLA
Model weights
Open
License
MIT
Parameters
1.2B
Pricing
Self-hosted
Benchmarks
1
Overall rank
Not ranked
Needs a simulation result to rank overall.
Real robots rank
#42of 56
Rating 1361
Simulation rank
Not ranked
No simulation results yet.
Rating
1338
10601989
Cost / task · est.
$0.239
$0.0087$17.74
Time / task
4min22s
9.5 s1 h 22 min
Rank on each benchmark
BenchmarkScoreRank
RoboChallengeProgress28.816 / 22

Per task attempt; the tiles show the median across benchmarks. Self-hosted compute is priced at one H100 at $3.29 per hour (Lambda on-demand, Oct 2026).

Other models from Tsinghua

Inference

Project page: 6 chunks/s (381 actions/s) with 64-action chunks, GPU unstated. OFT paper measured 297 ms latency on A100 (ALOHA, 3 images).