Development of deep reinforcement learning algorithms for humanoid robot to walk by avoiding obstacles

The operations to be performed by humanoid robots are considered to be a critically difficult task due to the impact of environmental factors. This thesis describes a possible control approach to increase the operational capabilities of humanoid robots to ensure target achievement in environments with various obstacles and walls. The humanoid robot has been trained using deep reinforcement learning methods based on the information received from the environmental elements it sees, the current location and direction. In the environments where the training takes place, one or more targets are given to the robot. The designed control approaches have been tested in Robot Operating System (ROS) and Webots simulation environments.