Adaptive and robust leg control of a hexapod robot for space exploration

Legged robots have shown incredible capabilities to operate on difficult terrains. This characteristic could be very useful for space exploration, but in this environment all the systems have to be very reliable and robust. The greatest challenge is to control the high number of degrees of freedom of the legs. To build a versatile and robust control system, it is crucial to have a single leg controller that is adaptive and versatile. The designed system has to be able to change the frequency and the length of its steps according to some controlling inputs. To accomplish this task using a low computational power and at the same time maintaining a high robustness, the design of the control system is biologically inspired from studies on the behavior of stick insects. This approach leads to the development of a dynamic artificial neural network. After a training phase, this system is able to obtain a walk with all the required behaviors. In addition, even though it receives data from the sensors to adjust its behavior to the external environment, it is not strictly dependent on these feedbacks, and it can even work in case of sensors failures. At the end, the obtained system is a simple pre-trained network that can activilely adapt its behavior, while being highly robust. Thanks to these qualities, this controller could be the optimal basis to build a complete control system for a hexapod robot in future works.