Theoretical Framework and Experimental Validation of Neuromorphic Computing Architectures for Real-Time Control of Compliant Robotic Manipulators

Compliant robotic manipulators with series elastic actuators and variable impedance joints, designed for safe physical human-robot interaction in Nigerian assistive technology and collaborative manufacturing applications, impose extreme real-time control demands involving simultaneous multi-contact force regulation, impedance shaping, and rapid reactivity to unstructured interaction events that challenge the processing latency and energy efficiency of von Neumann computing architectures conventionally used for robot control. This dissertation develops a theoretical framework and experimental validation methodology for neuromorphic computing-based real-time control of compliant robotic manipulators, exploiting the event-driven, massively parallel, and energy-sparse computation characteristics of spiking neural network hardware. A spiking neural network formulation for compliant manipulator impedance control is derived by translating the continuous-time impedance control law into a population coding and temporal coding framework compatible with spike-based computation. A theoretical analysis of spike timing-dependent plasticity as an online adaptation mechanism for impedance parameter tuning during unstructured interaction is conducted, yielding closed-form convergence conditions for the plasticity rule parameters. Implementation of the proposed SNN controller on an Intel Loihi 2 neuromorphic chip demonstrates 94 percent reduction in control loop power consumption and 62 percent reduction in control cycle latency compared to an equivalent GPU-accelerated deep neural network controller. Experimental validation is conducted on a five-degree-of-freedom compliant manipulator prototype performing peg-in-hole assembly and surface following tasks in both structured and unstructured contact scenarios, achieving force tracking accuracy within 0.8 Newtons. Keywords: neuromorphic computing, spiking neural network, compliant robot, impedance control, real-time control