Robot Muscles: Biologically-Inspired Cellular Actuators Using Distribute Stochastic Control with Broadcast Feedback
Speaker: Prof. Harry Asada , MIT ME
Muscle-like actuators having high energy-density and high-force, low-speed characteristics matched to the load impedance are expected to revolutionize robot design. Humanoids and other robotic systems recently developed have an increasing number of degrees of freedom, needing novel actuators replacing traditional electromechanical derives. In this seminar, recent progress in robot actuators using PZT, SMA, and conducting polymers will be presented. These actuator materials, although an order-of-magnitude higher in stress and energy density, are difficult to use due to prominent hysteresis, creep, and limited life cycle. To cope with these difficulties, novel system design and control methodology based on biologically-inspired cellular architecture will be presented in this seminar. A muscle-like actuator structure comprising a vast number of cellular units, each taking bi-stable ON-OFF states, will be developed and implemented. To coordinate a vast number of cellular actuators, a new control method, called “broadcast feedback”, will be developed for distributed stochastic control of numerous cellular units. In broadcast feedback, only a few global output signals are “broadcasted” to all the cellular units, which in turn make a probabilistic decision depending on the broadcast information and local state observation. Although there is no deterministic coordination among the vast number of cellular units, the ensemble of the cellular actuators can track a given trajectory accurately and robustly. Although 30 percent of the cellular units are dead, the system can still track the trajectory. The cellular architecture will be applied to a five-fingered humanoid hand and a snake robot for aircraft assembly.