Autonomous Motor Skills for Versatile Dynamic Legged Locomotion in Realistic Scenarios

The project is aiming to optimise the locomotion capabilities of legged robots using the state-of-art machine learning technologies. The uses of such technologies allow the legged robot to learn through the most efficient methods, which in conclusion will aid with locomotive application in order to enhance actions, gaits and overall stability. Inducing deep reinforcement learning algorithms into the control system also improves the dynamic interaction between the coexisting components of robots in parallel with the surrounding environment in order to refine and maintain a stabilised state supporting a desirable output. The systematically generated intelligent decisions will be adopted through deep neural networks in order to create memory-based functions, which can be optimised off-line and furthermore allow physical robots to inherit them under realistic scenarios and environments. Subsequently, the vision of the use of AI within the robots will not just aid with the motion control within the constitutional system but also induce a grasp of identifying the surrounding area to detect objects and terrain to further supplement the gait in order to maintain locomotive efficiency.