← Back to Open Data
verifiedsource · paperenv · BridgeData V2 + Open-X eval suite

Franka Research 3×openvla-7b

Franka Robotics · PyTorch · openvla

Evidence

Success rate
77.0%
Episodes tested
200
Updated
2026-04-20

https://arxiv.org/abs/2406.09246

Notes

OpenVLA-7B fine-tuned from Llama-2 + DINOv2 + SigLIP, evaluated zero-shot on BridgeData V2's Franka split. Strong baseline for cross-embodiment manipulation; the paper explicitly tests transfer to a Franka Research arm distinct from the training mix.

Gaps noted

  • · Numbers from the OpenVLA paper's BridgeData V2 evaluation slice — generalization to non-Bridge tasks is not measured here.
  • · No language-grounded long-horizon evaluation; success rate aggregates short-horizon pick / push / open primitives.

Robot · Franka Research 3

Franka Research 3, 7-DoF torque-controlled research arm. Successor to Franka Emika Panda. De facto manipulation-research standard.

View robot profile →

Policy · openvla-7b

Policy model from openvla. Tags: transformers, safetensors, openvla, feature-extraction, robotics.

View policy profile →