← Leaderboard

TraceVLA

VLA
University of Maryland · Dec 13, 2024 · built on OpenVLA

OpenVLA fine-tuned on 150K trajectories annotated with visual traces, which are drawn on the input image to give the model spatial-temporal context (University of Maryland with Microsoft Research). A 4B TraceVLA-Phi3 variant was also trained. Its SimplerEnv results cover the Google Robot tasks only.

Type
VLA
Parameters
7B
Backbone
Prismatic (Llama 2 7B)
Weights
Open
License
MIT
Commercial use
Allowed
Released
Dec 13, 2024
Robots
Google Robot (sim), WidowX 250, Franka Panda (sim)
SO-101
–
LeRobot
–
Inputs
RGB · L
VRAM
–

Benchmark results

1 of 13 benchmarks
LIBEROSaturatedSimulation
#68 of 7074.8%, self-reported by the authors, not a leaderboard entry
Not reported on 12 benchmarks
Wrong or missing number? Report an error