← Leaderboard

RoboCat

Policy
Google DeepMind · Jun 20, 2023

Multi-embodiment, multi-task generalist agent based on Gato that adapts to new tasks and robots from 100–1,000 demonstrations and generates its own training data for later iterations. Results are self-reported on DeepMind’s own sim and real tasks (on unseen tasks with 500 demonstrations each, success rose from 36% for the first version to 74% for the final one), not on a public benchmark.

Type
Policy
Parameters
—
Backbone
Gato architecture (visual goal-conditioned decision transformer)
Weights
Closed
License
–
Commercial use
Unclear
Released
Jun 20, 2023
Robots
real arms with parallel grippers, arm with three-finger gripper, simulated arms
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