Self-Driving Cars: Hands-On with Duckietown

This introductory module familiarizes students with the course logistics, navigation interface, and resources. It sets the stage for the learning journey in the world of robot autonomy with Duckietown.

This module introduces the potentials and challenges of self-driving cars, levels of autonomy, and the vision for autonomous vehicles. Students will set up their learning environment and Duckiebot, and make their first challenge submission.

This module covers the basics of making a robot, sensorimotor architectures, stateful architectures, and logical and physical architectures. Students will create a reactive "Braitenberg" agent to avoid duckies and compare their agent to other submissions.

This module introduces control systems, representations and models, and PID control. Students will design an odometry function and PID controller to command their Duckiebot's angular velocity.

This module covers projective geometry, camera modeling and calibration, and image processing. Students will develop image processing techniques necessary for visual lane servoing to control their Duckiebot to drive within markings.

This module introduces neural networks, convolutional neural networks, and one and two-stage object detection. Students will train a CNN to detect duckies and integrate their model with ROS to run onboard their Duckiebot for obstacle avoidance.

This module covers the Bayes filtering framework, parameterized methods (Kalman filter), and sampling-based methods (particle and histogram filter). Students will build a state estimation algorithm to predict their Duckiebot's pose as it travels.

This module introduces Markov decision processes, value functions, policy gradients, and domain randomization. Students will explore the capabilities and limitations of reinforcement learning models when applied to real-world robotics tasks such as lane following.