머신러닝과 립모션을 이용한 모바일 로봇의 동작 제어 및 자율 주행 시스템

Abstract

In this thesis we propose an autonomous navigation system that can control the motion of mobile robots using machine learning and leap motion. Robots have traditionally operated by remote control and have been applied to various industries. Especially, robots have operated at dangerous environments such as high, low temperature and space where are considered to unsuitable for humans, and very simple and repetitive tasks. As the application fields of robots have been expanded, the intelligence of robot is accelerated. However, due to the various factors including the limitation of artificial intelligence and uncertainty of work environment there is a limit to the complete intelligence of robot. Researches for machine learning have been carried out steadily to solve these problems and the current level of machine learning was recently confirmed through the Go competition of AlphaGo and Sedol Lee. Machine algorithm used in this thesis utilizes Q-learning, one of reinforcement learning algorithms, which is a learning method to achieve the desired goal by choosing the behaviors to be performed according to the current environment without the system model. The Q-learning was implemented using Tensor Flow of Google’s open software. Mobile robot with Q-learning algorithm can find the optimal path and also avoid dynamic obstacles using an obstacle avoidance algorithm. Mobile robot used in this thesis has one manipulator to carry out given specific missions. Leap motion used to control the manipulator can acquire 3D information of hand motion. The leap motion device uses the infrared camera and LED to recognize hand motion 200 times per second and to distinguish changes of hand motion up to 0.01mm. The effectiveness of the proposed system was verified through the experiments of autonomous navigation using machine learning and manipulator control using leap motion.