|Abstract: ||Humans can interact with their environment by tuning the properties of the musculoskeletal system to control force and impedance at the contact points with the environment. Many activities that require the use of tools, such as handling a screwdriver or chisel, involve an unstable interaction with the environment. This instability will amplify any disturbance or noise and cause unpredictable motion. For example, when chiselling a piece of rough wood, motor noise or involuntary movements can make the hand slip out of the desired track. When a single reaching movement is repeated in unstable dynamics, the central nervous system (CNS) learns to compensate for the instability by coordinating muscle in an appropriate way. However, it is still unclear how humans can learn to perform unstable tasks in various directions. Do they co-contract in an isotropic way to ensure good performance in all directions, or do they learn direction specific impedance? And what are the underlying processes of this adaptation?
This thesis addresses those questions through psychophysical experiments in which arm movements were investigated as subjects interacted with computer controlled dynamics produced by a robotic interface, as well as by developing computational models of human motor learning. Observations of learning unstable dynamics simultaneously in two different directions show that the CNS is able to compensate for the instability specifically for each direction, by adapting impedance optimally to the environment.
A computational model was developed to simulate human reaching movements in stable and unstable dynamics in various directions of the workspace. The model learns feedforward muscle tension by minimising movement error in muscle space and energy. The computer model is a useful tool to predict and investigate (generalisation in) motor learning. It can predict the results of our experiment as well as many motor learning observations in past experiments.|