Viscosity-based Height Reflex for Quadrupedal Locomotion on Rough Terrain

Legged robots are meant to traverse unstructured environments. In some cases, high steps (level changes) can pose significant challenges. These terrain features can be successfully addressed by whole body optimization when a map of the environment is available. Despite recent advances in this field [8], optimal planning is still far from being realized in real time, due to the complexity of the optimization process, the need to have both reasonably accurate 3D maps of the environment [5] and reliable state estimation [3]. In particular these approaches are of limited applicability in cases of visual deprivation where a reactive approach is preferable [1, 6]. In this abstract we propose a height reflex strategy to successfully negotiate big height changes (e.g. when the robot steps down from a high pallet). In this context, an important issue is the loss of mobility of the swing leg, when trying to establish a new foothold during a reaching motion [7]. An elegant way to address this situation, is to ”squat” the whole trunk to aid the reaching motion. In other words, the swing motion can be ”reconfigured” by mapping it among the four legs. This results in a reaching motion without stretching too much the swing leg. Another way of seeing this is that the height reflex is able to extend the workspace of the swing leg, through the motion of the body. This behaviour is somewhat similar to the one presented in [2], but we tackle the problem from a planning perspective rather than a control one. With our approach we were able to address higher obstacles, up to 24cm which is 33% of the leg length and 58% of its retractable leg range [9]. The proposed approach is also able to incorporate kinematic limits.