rldx1-ft-simpler-widowx-nf4

Automate WidowX robot manipulation tasks by running a fine-tuned RLDX-1 vision-language-action policy on the SIMPLER benchmark. Execute the openral command with action, object, and scene parameters to perform pick-and-place actions in tabletop environments.

A vision-language-action policy maps visual observations and language instructions to robotic actions. RLDX-1 uses this approach to enable precise manipulation, allowing robots to perform tasks like picking up objects and placing them on surfaces based on scene context.

Yes, the OpenRAL runtime is required to execute this fine-tuned RLDX-1 policy. You also need PyTorch and Transformers installed, along with specific hardware or a simulation setup matching the WidowX embodiment for proper operation.

No, this WidowX manipulation policy is designed for non-commercial use cases only. It is fine-tuned on the SIMPLER WidowX benchmark for research and evaluation, specifically ensuring effective real-to-sim correlation for the Bridge protocol.

Yes, this skill supports simulation environments alongside real-world WidowX robot setups. It is benchmarked on SIMPLER WidowX, which ensures effective real-to-sim correlation for evaluating pick-and-place tasks in tabletop scenes.

RLDX-1 for tabletop manipulation is limited to WidowX embodiments and non-commercial use. It requires specific hardware or simulation setup, plus dependencies like PyTorch and Transformers, restricting deployment to compatible environments.