AMAM 2008 Abstracts

Most human infants, before standing and walking on two feet, locomote on their four limbs. They can use a wide variety of locomotor patterns, but the most widespread one is alternated locomotion on hands and knees. Despite being the first locomotor pattern of humans, there are almost no quantitative information available on infants crawling. Since infants crawl only for a short time period of their life, use only one gait and have a very different functional limb geometry compared to other quadruped mammals (indeed infants use two functional segments for their fore limbs and one for the hind limbs since they crawl on their hands and knees), they offer an interesting way of studying quadruped locomotion in a simple form in order to understand its important characteristics. In this contribution, we present results on quantitative analysis of infants crawling compared to other quadruped mammals. Then, based on these observations, we formulate a set of desirable properties for the controller of a quadruped robot. Finally, we present such a controller for different robots, especially for an infant-like humanoid robot that crawls on its hands on knees, the iCub [1]. The controller design is based on central pattern generators (CPGs) constructed as networks of coupled oscillators. We also show how sensory feedback for online modulation of the control policies can improve the robustness of locomotion. Through kinematic analysis of several crawling infants, we show