A framework for whole body augmentative exoskeleton control

In this thesis, I present two primary contributions towards more capable augmentative exoskeleton systems including (1) the design and implementation of a robot-agnostic, high-level control infrastructure for better real-time performance and (2) a cohesive framework for whole-body augmentative exoskeleton control in a high-degree-of-freedom (dof) exoskeleton system. Both contributions were part of a larger project, in which our team designed and built a form-fitting lower-body augmentative exoskeleton with the objective to enhance a pilot's load carrying ability without sacrificing speed or maneuverability. Modern high-level control systems require excellent timing and low communication latencies to ensure stable, robust, and high-performance multijoint control. Towards this end, I designed and implemented a nodelet-based high-level controller wrapper that abstracts away and optimizes many of the implementation details involved in building such a control infrastructure. My first iteration of improvements used ROS's (Robot Operating System) intraprocess communication protocol along with proper integration of our RT-preempt kernel to ensure reliable, low-jitter timing performance and low-latency communication. I then helped to design infrastructure improvements that further reduced round-trip times via full system synchronization. My high-level control infrastructure has enabled significant advances for a variety of projects in the Human-Centered Robotics Lab (HCRL), including the development of a controller for an augmentative exoskeleton and dynamic walking using the lab's point-foot bipedal robot, Mercury. The second major contribution in this thesis is an algorithm that I developed for whole-body augmentative exoskeleton control. It uses a model of the exoskeleton to cancel static, gravitational loads, and measured cuff forces to attenuate human-exo interaction forces, including inertial loads and those caused by disturbances from the environment. The key contribution of this control scheme relative to other exoskeleton transparency controllers is how this algorithm (1) handles contact switching given the corresponding discrete changes in the dynamics and (2) routes the needed reaction forces to the ground given the underactuated, floating-base dynamics with contact constraints. I formulate a Quadratic Programming (QP) optimization problem to solve for permissible reaction forces and actuator torques that come as close as possible to providing the desired dynamic attenuation behavior of the controller while also satisfying wrench-cone constraints for each of the exoskeleton's contacts. A relaxation variable, penalized in the cost function, ensures the solver can always find a feasible solution, and cost function weights penalizing contact point accelerations and reaction force magnitudes are smoothly interpolated to ensure continuous torque commands as the system switches between two discrete sets of contactsMechanical Engineerin