Introduction
I have always had a passion for robotics research. However, it took several years and multiple detours to realize that my true passion for robotics is in controls - making robots move in an interesting and efficient way. Coming from a hardware background (i.e. mechanical engineering), I have a unique perspective that I believe could contribute to software engineering world of robotics. While I attempt to transition specialty from mechatronics to software/ML within robotics, I have been teaching myself Tensorflow and learning about the latest developments and current trends in AI and Robotics. Traditionally, there are three main topics in robotics, mapping, localization and motion control. The most popular method for navigation is simultaneous localization and mapping (SLAM), which can be thought of as a classical AI method using statistical techniques like particle filter or Kalman filter. Recent advancements in convolutional neural net and computer vision have created a field of perception in robotics where sensory information is used to make higher-level decisions in motion planning. In this series, I summarize the most interesting findings in the application of machine learning for robot manipulation through select examples in the literature. I will first survey the current methods of teaching robot manipulation, followed by a discussion of the applications of specific deep reinforcement learning, imitation learning, or meta-learning algorithms for robot controls.