Robot Learning with Minimal Human Feedback

Explore a cutting-edge robotics colloquium featuring Erdem Biyik from USC, focusing on robot learning with minimal human feedback. Delve into innovative techniques that enable training robots using limited human input, such as single demonstrations, language instructions, or natural eye gaze. Discover the challenges of collecting large robotics datasets and learn about alternative approaches to improve data efficiency in robot learning. Gain insights into the application of reinforcement learning from human feedback and the use of language corrections as an efficient training method. Examine how existing large pretrained vision-language models can generate direct supervision for robot learning. Understand the potential impact of these techniques on overcoming obstacles in robot learning and advancing the field of robotics.