| The design of the control and hardware systems for a robot intended for interaction with a human user can profit from a critical analysis of the human neuromotor system and human biomechanics. The primary observation to be made about the human control and "hardware" systems is that they work well together, perhaps because they were designed for each other. Despite the limited force production and elasticity of muscle, and despite slow information transmission, the sensorimotor system is adept at an impressive range of motor behaviors. In this thesis I present three explorations on the manners in which the human and hardware systems work together, hoping to inform the design of robots suitable for human-robot interaction.;First, I used the serial reaction time (SRT) task with cuing from lights and motorized keys to assess the relative contribution of visual and haptic stimuli to the formation of motor and perceptual memories. Motorized keys were used to deliver brief pulse-like displacements to the resting fingers, with the expectation that the proximity and similarity of these cues to the response motor actions (finger-activated key-presses) would strengthen the motor memory trace in particular. Error rate results demonstrate that haptic cues promote motor learning over perceptual learning.;The second exploration involves the design of an actuator specialized for human-robot interaction. Like muscle, it features series elasticity and thus displays good backdrivability. The elasticity arises from the use of a compressible fluid while hinged rigid plates are used to convert fluid power into mechanical power. I also propose impedance control with dynamics compensation to further reduce the driving-point impedance. The controller is robust to all kinds of uncertainties.;The third exploration involves human control in interaction with the environment. I propose a framework that accommodates delays and does not require an explicit model of the musculoskeletal system and environment. Instead, loads from the biomechanics and coupled environment are estimated using the relationship between the motor command and its responses. Delays inherent in sensory feedback are accommodated by taking the form of the Smith predictor. Agreements between simulation results and empirical movements suggests that the framework is viable. |
