MantisBot Changes Stepping Speed by Entraining CPGs to Positive Velocity Feedback

This paper demonstrates and analyzes how CPGs can entrain joints of a praying mantis robot (MantisBot) to positive velocity feedback resulting in a duration change of a leg’s stance phase. We use a model of a single leg segment, as well as previously presented design techniques to understand how the gain of positive velocity feedback to the CPGs should be modulated to successfully implement the active reaction (AR) during walking. Our results suggest that the AR simplifies the descending control of walking speed, naturally producing the asymmetrical changes in stance and swing phase duration seen in walking animals. We implement the AR in neural circuits of a dynamic network that control leg joints of MantisBot, and experiments confirm that the robot modulates its walking speed as the simple model predicted. Aggregating the data from hundreds of steps in different walking directions show that the robot changes speed by altering the duration of stance phase while swing phase remains unaffected, as seen in walking animals.