Research on Humanoid Robot Control Based on Environmental Feedback

This paper investigates systematically key technologies in order to make humanoid robots real time adaptive in dynamic unstructured environment. It builds up a closed loop architecture for development of control paradigms, hardware configurations physical limitations, and state reconstruction of sensing systems. At the control level, this research focuses on the combination of model-based optimal control and data-driven learning approaches, where a hybrid architecture is proposed where model-based safety boundaries are used for learning. With regard to the hardware and sensing, the significance of morphological computation in sharing of control task is revealed from this research and the transition of sensor configuration towards multi-source heterogeneous fusion of vision, inertia and touch to resolve state loss in complex conditions is discussed. In addition, this research stress on the point that future developments will be limited to 'biological fusion' and bringing together algorithmic 'transparency' and generalization to enable robots to advance from simple locomotion to high level cognitive interaction. This is the conclusion proposed by this research, which suggests that future technological advances will come from the efficient synergy of perception, control and hardware, and hardware and software co-design, which will enable robots to truly move on to the open world.