← Leaderboard

GR-MG

Policy
ByteDance Research · Aug 26, 2024

Policy conditioned on both a language instruction and a generated goal image, so it can learn from partially annotated data. Real-robot results are self-reported over 58 tasks (success rising from 68.7% to 78.1% in the simple setting); it also reports CALVIN.

Type
Policy
Parameters
—
Backbone
GR-1-style policy + InstructPix2Pix goal-image generator
Weights
Open
License
Apache-2.0
Commercial use
Allowed
Released
Aug 26, 2024
Robots
Kinova Gen3, Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
–
VRAM
–

Benchmark results

1 of 13 benchmarks
CALVIN ABC→DNear saturationSimulation
#15 of 244.04 / 5, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error