Machine Learning applied in the modern robot

Implemented some of the most popular used machine learning algorithms and applied them into the real robot cases.

1. Object Detection and Geometry distance estimation based on GMM: trained some GMMs based on the image color data, also combined image processing and geometry algorithms to detetct any read barrel in a robot’s view and estimate its real geometry distance from robot. source code
2. Robot Gesture Recoginition using Hidden Markov Model (HMM): built an algorithm based on the HMM to recognize different robot arm motion gestures. Finally, it was able to classify unknown arm motions within almost real-time. source code
3. Reinforcement Learning using Policy Gradient Methods: implemented several methods for dealing with the Robot Walking on the Frozen Lake problem. The situation can be represented via a Markov Decision Process(MDP) and a strategy for retrieving the frisbee can be obtained using value iteration (VI), policy iteration (PI), and policy gradient optimization (PGO). source code