Episode Details

MotorMind leverages a frozen VLM with mid-level prospective actions and asynchronous feedback to control robots zero-shot on manipulation tasks, requiring no task-specific training. This advances robot foundation models by demonstrating strong generalization across manipulation benchmarks without fine-tuning.