Final Report
Joshua Davy
May 2020
Traditionally the gaits of legged robots are inspired by what is seen in nature and on assumptions of what properties the optimum gait would have. This involves kinematic and dynamic analysis of the robot which becomes increasingly difficult with robots with greater degrees of freedom. Further, real actuators and joints are not ideal and are also difficult to mathematically model accurately. In this project the aim is to use optimisation techniques to allow a robot to develop its own gait free from human bias of what the optimum gait would look like. This technique means no kinematic or dynamic analysis of a robot is necessary, as the robot will attempt to learn a policy of moving its joints to optimise its gait. The project will consist of two main sections, firstly the design and build of a suitable robotic platform for experimentation and secondly, the investigation and design of suitable software to control the robotic platform.