Your robot can now sense, plan, and navigate. This last module is about making it learn and about testing it safely before it ever touches real hardware. Machine learning lets a robot improve from data and experience instead of relying only on rules you wrote by hand.
You'll start with neural networks, the function approximators behind modern perception and control. Then you'll meet the Markov decision process, the mathematical framework that underlies reinforcement learning, where a robot learns behavior by trial and reward.
Learning from trial and error is slow and risky on a real robot, so you'll finish with simulation. You'll run robots in Gazebo with ROS, and see how MuJoCo provides fast, accurate physics for training and research. Together these close the loop on everything in the course, and they set you up for the Projects.
A short introduction to neural networks, the building block behind most modern machine learning used in robotics.
The framework behind reinforcement learning: states, actions, rewards, and how an agent chooses what to do.
How to spawn a robot in Gazebo, control its joints, build a virtual world, and add a camera sensor using ROS.
An overview of MuJoCo, the physics simulator widely used for robot learning and reinforcement learning research.