DISC: Decoupling Instruction from State-Conditioned Control via Policy Generation
Researchers have developed a new method called DISC (Decoupling Instruction from State-Conditioned Control) to improve language-conditioned manipulation policies in robotics. DISC structurally separates instruction processing from state-conditioned control, preventing policies from learning shortcuts that bypass language grounding. It achieves this by using a hypernetwork to generate task-specific visuomotor policies directly from instructions, ensuring task awareness is derived solely from language. AI
IMPACT This method could lead to more robust and adaptable robotic systems that better understand and execute complex instructions.