Adaptive Human Model-Based Control for Active Knee Prosthetics

Every unimpaired person in the world walks thousands of steps every day, and has an individual gait pattern, e.g. different velocities, load conditions, surrounding/ground conditions. Unfortunately, the number of lower-limb amputees has been increasing during the last centuries and millions of people are currently affected by amputation. To restore human walking abilities, an active knee prosthesis needs to be seamlessly integrated into the human control-loop. The aim of this thesis is to design a biomimetic adaptive impedance control for active knee prostheses. This controller has to be capable, firstly, of reproducing the physiological impedance modulation of a healthy leg, secondly, of adapting the impedance of the leg depending on its interaction with the environment. To determine essential characteristics of the control-loop, physiological models of human-based control were considered, and design requirements were derived from biology. To design the biomimetic adaptive controller, two different impedance models have been tested, i.e. a spring damper model (BK) and a neuromuscular skeletal model (NMS). Then, a Human Model Reference Adaptive Control (HMRAC) has been tested, with both impedance models, using different High Level Controllers (HLC), i.e. Complementary Limb Motion Estimation (CLME) and Finite State Machine (FSM). The HMRAC modifies the knee impedance depending on the error in position between the actual knee angle and a reference knee angle estimated online from the user via CLME. When the HMRAC is active, the error in position can be minimized to 4.44 °, which is below the smallest perceivable error in position for human lower limbs. The torques generated respect the profile and the range of the human physiological torques. Three healthy test subjects reported an improvement in terms of comfort and perception when the HMRAC control with NMS impedance model was used. On one hand, these results show that the NMS impedance model is superior to the BK impedance model if an appropriate set of user-specific impedance parameters is used. On the other hand, the HMRAC is able to modulate the impedance of the leg in a physiological way: the knee torque is changed with the minimum effort to reduce the error in position w.r.t. a desired trajectory estimated from the user intention.