Force-based control of a six-legged robot on abrupt terrain using the subsumption architecture

We have developed a controller for a six-legged robot that allows it to walk on rough or even abrupt terrain using only force sensors as feedback. In an initial ap-proach, which at rst seemed natural, we took a subsumption-based controller tailored to walk on at surfaces and tried to add a force-compliance layer on top of it, thus maintaining the modularity of the design and building the controller in a purely in-cremental way, as proposed by Brooks in [1]. However, this approach could not be followed without making substantial modications in the existing layers, in contra-diction with our requirement of incrementality. Modularity was also lost, since lower levels of competence depended on the upper ones to work. A solution to these prob-lems was found by redesigning the layer structure into a new one in which (compliant) walking corresponds to the uppermost level of competence, and in which a level for non-compliant walking has been ruled out.