Deep Reinforcement Learning for Walking Robots

Sebastian Castro demonstrates an application of deep reinforcement learning in controlling humanoid robot locomotion using the Deep Deterministic Policy Gradient (DDPG) algorithm. The robot is simulated using Simscape Multibody™, while training of the deep reinforcement learning policy is done using Reinforcement Learning Toolbox™.

The video outlines the setup, training, and evaluation workflow of deep reinforcement learning. First, Sebastian introduces how to choose states, actions, and a reward function for the deep reinforcement learning problem. Then he describes the neural network structure and training algorithm parameters. Finally, he shows the training results and discusses the benefits and drawbacks of deep reinforcement learning.

You can find the example models used in this video in the MATLAB Central File Exchange.

For more information, you can access the following resources:

• Reinforcement Learning Toolbox
• Reinforcement Learning Tech Talks
• Blog and Videos: Walking Robot Modeling and Simulation
• Paper: Continuous Control with Deep Reinforcement Learning
• Paper: Emergence of Locomotion Behaviours in Rich Environments