Modular, multimodal arm control models

Abstract

Human and animal behavior can be amazingly flexible and adaptive. Even when only considering the dexterity of human arm movements, a rather complex control architecture appears necessary. This control architecture faces three particular challenges, which we discuss in detail. First, sensory redundancy requires the flexible consideration, combination, and integration of different sources of information about the state of the arm and the surrounding environment. Second, motor redundancy requires the flexible consideration and resolution of behavioral alternatives. Third, the continuous uncertainty about body and environment requires flexible control strategies that take these uncertainties into account. Research in cognitive modeling as well as in psychology and neuroscience suggests that the human control system effectively solves and even partially exploits these challenges to generate the observable dexterity. Besides theoretical considerations from control and cognitive modeling perspectives, we survey the capabilities and current drawbacks of the sensorimotor redundancy resolving architecture (SURE REACH) of human arm reaching. Moreover, we consider an even more modular model of human motor control, which is currently being developed. Both architectures can yield the dexterous behavioral control observable in humans, but only the latter scales to many degrees of freedom. Thus, the architectures may provide insights on how dexterous motor control is realized in humans and on howmore adaptive and flexible robot control systems may be developed in the future.