From robotics to latest deep RL algorithms implementation, my portfolio of recent projects is presented below.
You can click on the button below for understanding underlying theoretical concepts used in the projects
Evidence suggest our perception is governed by our brain's prediction of the future based on our internal model. Can we build models of the world? Can we train and learn in our imagined world? This project is the reproduction of the paper, Ha and Schmidhuber, "Recurrent World Models Facilitate Policy Evolution", 2018
Until recently, the Go Game was considered as untractable for computers because of its huge search space. This is not the case anymore since the defeat of a Grand Master, Lee Sedol, in 2016. The project is the implementation of AlphaZero algorithm (applied to the TicTacToe game for reduced computation), based on Deepmind's paper, Mastering the game of Go without human knowledge, published in October 2017
Despite the sample-efficiency challenges, deep Reinforcement Learning (RL) techniques are increasingly showing successes as demonstrated by DeepMind's achievements since 2015 (Atari game to AlphaFold). Many approaches exist in deep RL: value-based and policy-based methods, model-free and model-based, on-line and offline learning or single-agent and multi-agent settings. The repository contains a suite of projects and tutorials that cover a wide range of current techniques in deep RL.
Self-driving cars, drones and autonomous robots need to understand their current state in the environment (e.g. localization), search and plan in complex environments (e.g. an optimal path in a urban environement), and then control its actuators (e.g. wheel or propeller) for executing their objective. The repository includes 11 projects addressing these real challenges (2D and 3D)
Please feel free to email me and leave comments about my work. I look forward to hearing from you