Optimal Robotic Leg Design: Performance Metrics for an Optimization Framework

Abstract

This dissertation presents a framework for simultaneous optimization of motors, transmissions, and mechanisms across the joints of robotic legs, with the goal of achieving energy efficient, precisely controllable, and stable locomotion in dynamic environments. Within this framework, two novel performance metrics are introduced for separate evaluation of stability during stance and swing. The metrics are first tested in simulation with a reduced order model, which motivates an extended forced-oscillation based reduced-order model that includes an ankle joint and a finite sized foot. The model aligns more closely with human-like gait characteristics, and its human-inspired ankle push off strategy maintains stability against small perturbations, as observed in humans. A case study demonstrates the importance of considering the two stability metrics together with an energy metric. The study is extended to a multi objective constrained optimization for a 3-D leg mechanism, yielding an anthropomorphic articulated leg with 6 actuated degrees of freedom and an optimal actuator design. Using this leg design, which has mass and inertia distributions similar to those of humans, a humanoid robot, Mithra, was designed and realized. Mithra is 1.75 m tall, has a total mass of 75 kg, and provides leg motion in the three principal planes. Its development included engineering analysis of structural components, sensor system development, real time communication, embedded programming for real time control and motor control, and battery management system development. Preliminary experiments indicate that Mithra achieves human-like kinematic and dynamic characteristics and provides sufficient actuator and structural strength for tasks such as stair ascent and descent, squatting, and running. Results also show performance that closely matches the elegance of human locomotion and outperform other state-of-the-art humanoid robots. The actuators of Mithra appear capable of supporting human inspired control paradigms and may enable the first electrically actuated humanoid to achieve human-like running at 3 m/s.