Humanoid robotics: What we really need - just machines, or something more?

摘要

We are witnessing research expansion and application of robotics in almost all segments of human's activity, from industry to space exploration. Similar happen with humanoid robotics, but it has to be pointed out that its applications areas are mostly situated in environment close to humans. This imposes some specificities that have to be encountered. Two of them are, in our opinion, of essential importance. First one is that human environment is either loosely structured or completely unstructured (as a consequence, walking realization by programming it in advance is not possible and on-line generation of walk is inevitable), while second one is that robot's physical contact with surrounding objects, humans or other robots almost can not be avoided. If unexpected (and unwanted) such contact can be considered as disturbance and can jeopardize humanoid's dynamic balance, whose preservation is prerequisite for any biped activity. In both situations new approaches are needed. We have proposed new method for on-line walking generation based on set of predefined primitive movements that can be easily combined into more complex ones. Thus, basic leg's motion is always same (combined of primitives), and just parameters of primitives (step length, height of foot trajectory, leg bending intensity, walking velocity, ...) have to be adjusted on-line during walking. The second issue, closely related to walking, is preservation of dynamic balance, either in static posture or during walking. Because existence of dynamic balance is assumed for any biped performing gait, its preservation is of primary importance. However, compensation strategies are not same for disturbances of various intensities. For real application this is extremely important and complex issue, particularly in case of large disturbances. Compensation action must simultaneously ensure two, often contradictory, requirements (to prevent immediate biped overturn and to bring system into state which enables previous activity to be continued when compensation action is over) and we propose compensation to be composed as synergetic action of more DOFs simultaneously.