Developing and comparing various reinforcement learning algorithms on the Lego Mindstorm robots in attempt to simulate an autonomous vehicle on a small scale in an obstacle course
The obstacle course is an 8x4ft (subject to change) rectangle with start and end points on opposite sides. The course will be surrounded by walls. The course will contain static and variable obstacles. Static obstacles will remain in the same position after each test. Variable obstacles will be placed randomly within a set perimeter each test iteration. The course will be constructed on a flat surface.
The obstacles, walls, and goal of the rectangle will be defined by different color sticky-notes/tape (i.e., yellow = variable obstacles, red = static obstacles, green = goal, black = walls). The robot will use visual sensors to determine the colors of the sticky-notes/tape.
A Lego Mindstorm robot that shows measurable improvement after successive obstacle course driving tests by utilizing reinforcement learning algorithms.
- A data analysis/visualization of the performance improvement and obstacle course environment information obtained after each test.
- An improvement rate comparison of each reinforcement algorithm tested. (e.g., Temporal Difference Algorithm, Expectation Maximization Algorithm)
Machine Learning with a Lego Mindstorms Robot | Stanford
Reinforcement Learning on the Lego Mindstorms NXT Robot. Analysis and Implementation
Lego Mindstorms NXT and Q-Learning: a teaching approach for robotics and engineering