Neural networks usage for optimization of robot arm path

In the rapidly modernized world of the 21st century, robots are beginning to play a significant role. Looking back, robotic manipulators and contractors were the very first robots used in the industry. They have been used since the 1960s. Nowadays, when computers have evolved very dramatically by the technical possibilities, humans managed to implement self learning algorithms to robots in simulation environments and reality. In those environments, robots which are using self learning algorithms like reinforcement learning (RL), can learn how to interact with different environments. This master’s thesis will aim to implement one of the main goals - to apply the algorithm of deep enhanced neural network to optimize the robot arm of eight freedom nodes and achieve the goal. In order of this goal to be achieved, tasks have been set: • To find out the principles of operation of the robot arm. • To investigate the applications of neural networks in the field of optimization. • Train a custom neural network to optimize the path of robot arm movement. • Test a trained network. • Visually simulate the obtained results to see a clear result of the work.62 During this master’s thesis a simulation environment was created with Unity platform. For machine learning algorithm, the Unity ML plugin was used which has Tensorflow algorithm integrated inside. Experiments were made with 8 and 6 freedom knots in environment with obstacles and without. Markov – decision process algorithm was used for AI learning. Results are positive because learned arms had quite accurate learning. Robotic arm tries to reach a goal and do not touch obstacles. Robot tries to find optimal path from beginning to the goal. Also, there are always space for improvements. One of them could be to teach robot arm to move smoothly