Browsing by Subject "Identification for robust control"

Now showing 1 - 1 of 1

Control strategies for series elastic, multi-contact robots
(2019-09-16) Thomas, Gray Cortright; Sentis, Luis; Djurdjanovic, Dragan; Bakolas, Efstathios; Chen, Dongmei
As robots designed for physical interaction with humans---humanoids, exoskeletons and beyond---make their entrance into society, understanding the limitations of their interaction behavior will be key to their effective use. The state of the art method for allowing such systems to be both compliant and force sensitive is to introduce mechanical springs into the joints of these robots, making them "series elastic". But this complicates the control of these robots, making it hard to separate truth from optimism in what they will be able to accomplish using feedback control. Robots are programmed in hierarchical layers, and each layer makes assumptions about the layer below it. The planning layer assumes the plan will be followed. The whole body controller layer assumes the actuators will supply whatever torque it specifies. And the actuator control layer assumes the actuator behaves like a linear system. This dissertation studies the interfaces between these layers as they are influenced by the choice to include series elastic actuation, hoping to resolve the mismatch between assumptions and guarantees that arise from this choice. These questions lead it naturally to the lowest of the layers, where a new system identification system allows the actuator to assume a bounded uncertainty model. The dissertation then refines the insights from studying uncertain SEA models into a simpler model that explains the most important factors. It uses this to design SEA controllers that go beyond the traditional limits of passivity. These insights also apply to the problem of strength augmentation exoskeleton control. Factor of 3 amplification results are reported on a tethered, 12 degree of freedom, powered, lower body exoskeleton with four passive joints using a simplified version of the controller and a far more advanced whole body control framework. These ideas are introduced in the context of the authors's work with various testbeds and state of the art robots including a point foot biped, the DARPA virtual robotics challenge simulator, the NASA R5 Valkyrie Humanoid, and the Apptronik Sagittarius Lower Body Exoskeleton.