Implemented both feature-extraction based algorithms and end-to-end Deep Q learning method to learn to control objects in-game directly from high dimensional game screen input (using Opencv ). The DQN is a Convolutional neural network that reads in pixel data from the game and the game score.
Prototyping robots for PyBullet (F1/10 MIT Racecar, Sawyer, Baxter and Dobot arm, Boston Dynamics Atlas and Botlab environment)