← Leaderboard

GR-2

World-action
ByteDance Research · Oct 8, 2024

Video-language-action model pretrained on 38 million internet video clips and fine-tuned on robot trajectories to generate video and actions together. Results are self-reported on ByteDance’s own real-robot setup (97.7% average success over 105 tasks in the simple setting); its CALVIN result covers only the ABCD→D split.

Type
World-action
Parameters
230M
Backbone
GPT-style video-language-action transformer
Weights
Closed
License
–
Commercial use
Unclear
Released
Oct 8, 2024
Robots
Kinova Gen3
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