“Our benchmark consists of 100,000 tasks, including household chores such as cleaning up dishes and toys,” Meta writes. “We are also releasing the PARTNR dataset consisting of human demonstrations of the PARTNR tasks in simulation, which can be used for training embodied AI models.”
Simulation has become an increasingly useful tool in robot deployment, allowing organizations to test in seconds what might otherwise take hours or days to accomplish in the real world. Meta says, however, that it has also had success deploying the PARTNR model outside of simulation. It has already been used in Boston Dynamics’ Spot robot in testing. Meta has also built a mixed-reality interface designed to offer a visual representation of the robot’s decision-making processes.