Applying RL to Real-World Robotics with Abhishek Gupta - #466

This article summarizes a podcast episode featuring Abhishek Gupta, a PhD student at UC Berkeley's BAIR Lab. The discussion centers on applying Reinforcement Learning (RL) to real-world robotics. Key topics include reward supervision, learning reward functions from videos, the role of supervised experts, and the use of simulation for experiments and data collection. The episode also touches upon gradient surgery versus gradient sledgehammering and Gupta's ecological RL research, which examines human-robot interaction in real-world scenarios. The focus is on practical applications and scaling robotic learning.

• •Focus on applying Reinforcement Learning (RL) to real-world robotics.
• •Exploration of reward supervision and learning reward functions from videos.
• •Discussion of simulation for experiments and data collection.
• •Examination of gradient surgery vs. gradient sledgehammering.
• •Focus on ecological RL and human-robot interaction.