← Leaderboard

GR-1

Policy
ByteDance Research · Dec 20, 2023

GPT-style policy pretrained on large-scale video generation (Ego4D) and fine-tuned to predict both future frames and actions. On real robots its results are self-reported (0.79 object-transport success on seen objects vs 0.27 for RT-1); it also reports CALVIN.

Type
Policy
Parameters
195M
Backbone
GPT-style transformer + MAE ViT-B
Weights
Open
License
Apache-2.0
Commercial use
Allowed
Released
Dec 20, 2023
Robots
Kinova Gen2, Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
CALVIN ABC→DNear saturationSimulation
#21 of 243.06 / 5, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error