Sensory feedback and control of leg-substrate interactions in insects and robots

Sensory feedback plays an important role in adaptive control of insect leg movements as required, for example, in agile locomotion over irregular terrain. Functionally, insect legs are not just effectors--they are also sophisticated sensory devices that provide the nervous system with a great deal of information about the dynamic state of the leg and interactions with its environment. Most previous research on the neural control of multilegged locomotion has focused on the problem of pattern generation, namely on how to produce appropriately phased patterns of bursting activity in leg motor neurons so as to generate stable walking gaits. Fixed pattern generation alone, however, is not sufficient to achieve agile locomotion over irregular terrain.To investigate the role of sensory feedback from leg sense organs in movement control we have developed simulation software that allows us to study the dynamic interactions of all the elements in the sensory-motor loop: the neural controller, sensors, actuators, leg dynamics and interactions between the leg and its environment. We also develop and test ideas about sensory feedback in a physical robotic leg which incorporate many of the functionally important characteristics of insect legs. Each leg segment is equipped sensors that provide information similar to the ones found on insect legs. We modified the joint actuators to emulate the dynamics of antagonistic muscle pairs. Real-time neural controllers are implemented using digital signal processor technology.We describe neural controllers that implement a substrate-finding reflex in which the leg searches for and establishes a foothold on a spatially-restricted support. By analyzing relatively simple single leg reflexes, we demonstrate how rhythmic and adaptive behavior can appear as an emergent property of the coupled dynamics of the components of the sensory-motor loop. Using this platform as our testbed, we also explore the use of genetic algorithms and on-line learning algorithms to optimize the design of neural controllers.U of I OnlyETDs are only available to UIUC Users without author permissio