← Leaderboard

CrossFormer

VLA
UC Berkeley (RAIL) · Aug 21, 2024

Cross-embodiment transformer policy trained on 900K trajectories from 20 embodiments. One set of weights controls single- and dual-arm manipulators, wheeled robots, quadcopters and quadrupeds. It is evaluated only on the authors’ own real-robot setups, so it has no ranked scores.

Type
VLA
Parameters
130M
Backbone
Transformer trained from scratch (ResNet-26 image encoders)
Weights
Open
License
MIT
Commercial use
Allowed
Released
Aug 21, 2024
Robots
WidowX, Franka, ALOHA (bimanual), LoCoBot (navigation), Go1 quadruped, Tello quadcopter
SO-101
–
LeRobot
–
Inputs
RGB · P · L
VRAM
–

Benchmark results

0 of 13 benchmarks

No benchmark results recorded yet.

Not reported on 13 benchmarks
Wrong or missing number? Report an error