Coursera Flash Sale
40% Off Coursera Plus for 3 Months!
Grab it
Watch a 59-minute research presentation from MIT PhD candidate Felix Yanwei Wang exploring novel approaches for customizing pretrained robotic policies without additional training. Learn about two key frameworks - inference-time steering for single-step task specification through human interactions, and task and motion imitation using symbolic plans for multi-step specifications. Discover how framing policy errors as task mis-specifications rather than skill deficiencies enables more efficient deployment of generalist policies. Understand cutting-edge research in human-robot interaction and policy adaptation that has been recognized at major conferences like CoRL and ICLR, featured on PBS, and exhibited at the MIT Museum. Gain insights from Wang's work under Prof. Julie Shah at MIT and Prof. Dieter Fox at NVIDIA Robotics Lab on maximizing the utility of pretrained models while achieving user objectives during inference.
