OpenVLA-OFT
A vision-language-action (VLA) model that end-to-end combines large pretrained language and vision models with robot action prediction. It accepts image and natural language instruction inputs and directly outputs robot control action sequences.
Timeline (last 90 days)
The research team conducted tests on two vision-language-action (VLA) models, OpenVLA-OFT and π0.5, covering two benchmark environments, LIBERO-Spatial and LIBERO-Plus.
Under the τ=0.5 recovery criterion, among the 66 failure cases of OpenVLA-OFT on LIBERO-Plus, only 14 failures contained recoverable checkpoints using the 'continue execution' strategy alone, while allowing 'continue execution or scripted recovery' raised the number of recoverable cases to 34.
All Facts (2)
The research team conducted tests on two vision-language-action (VLA) models, OpenVLA-OFT and π0.5, covering two benchmark environments, LIBERO-Spatial and LIBERO-Plus.
50%UnverifiedUnder the τ=0.5 recovery criterion, among the 66 failure cases of OpenVLA-OFT on LIBERO-Plus, only 14 failures contained recoverable checkpoints using the 'continue execution' strategy alone, while allowing 'continue execution or scripted recovery' raised the number of recoverable cases to 34.
50%