Lecture
Reinforcement Learning for Robotics
- at -
- Hall C5C5.100
- Language: English
- Type: Workshop
Lecture description
"Reinforcement learning is reshaping how robots learn to move. With recent breakthroughs at labs like ETH Zurich, Boston Dynamics, and Disney Research, walking robots are now handling rough terrain and challenges they were never explicitly trained for. The good news is that you don't need a fully equipped research lab to get started. In this hands-on workshop, participants will train an AI agent from scratch to balance a tiny wheeled robot, then watch it run on real hardware.
Using the MuJoCo physics simulator and Python, we'll build a complete RL training pipeline using the Proximal Policy Optimization (PPO) algorithm. We'll tackle two of the most common real-world challenges in robotics: how to make an agent that's actually robust (using domain randomization and curriculum learning)... and how to transfer a policy learned in simulation onto a physical robot (a problem known as the ""sim-to-real"" gap). By the end of the session, attendees will leave with hands-on intuition for modern robot learning.
Target Audience: Embedded developers and makers with some programming experience. No prior robotics or RL knowledge required, though familiarity with Python, Arduino, and basic ML concepts will help participants get the most out of the session."
Using the MuJoCo physics simulator and Python, we'll build a complete RL training pipeline using the Proximal Policy Optimization (PPO) algorithm. We'll tackle two of the most common real-world challenges in robotics: how to make an agent that's actually robust (using domain randomization and curriculum learning)... and how to transfer a policy learned in simulation onto a physical robot (a problem known as the ""sim-to-real"" gap). By the end of the session, attendees will leave with hands-on intuition for modern robot learning.
Target Audience: Embedded developers and makers with some programming experience. No prior robotics or RL knowledge required, though familiarity with Python, Arduino, and basic ML concepts will help participants get the most out of the session."
