Application of Co-Design Principles for Design of Series Elastic Joints

Abstract

Compliant joints enhance the performance of dynamic legged robots by enabling more robust, efficient, and resilient locomotion. A widely adopted approach for introducing compliance into robotic joints is the use of Series Elastic Actuators (SEAs). Designing SEAs, however, requires balancing the stiffness of the elastic element with the structure and gains of the control system, as both strongly influence actuator bandwidth, disturbance rejection, and overall efficiency. Prior work across many domains has demonstrated that co-design methodologies, those that optimize mechanical and control parameters simultaneously, can produce high-performance, robust systems.

This dissertation advances the capabilities of dynamic legged robots through the development of a comprehensive co-design strategy for SEAs. The proposed framework addresses key limitations of traditional SEA design, particularly their difficulty in balancing the trade-off between high bandwidth achievable by stiff actuators and the disturbance rejection afforded by increased compliance. By jointly optimizing the gains of a simple, easily implemented PID–feedforward controller alongside the stiffness of the elastic element, the approach presented here improves both controllable bandwidth and transient response without requiring complex control architectures. A systematic method for identifying cost functions that are broadly applicable, implementation-friendly, and reliably indicative of system performance is presented. These cost functions are then used within a co-design optimization applied to several SEA configurations, demonstrating both generality and performance improvements over conventional sequential design approaches.

In addition, this work investigates how infill density influences the flexural rigidity of fused deposition modeling (FDM) printed PLA beams. These experiments support the use of FDM-printed components as compliant elements within SEAs. Using static three-point bending tests, regression models are developed to predict part flexural rigidity as a function of print infill. These models are integrated into the co-design framework, replacing direct selection of elastic stiffness with the specification of beam geometry and infill percentage. The resulting co-designed hardware is validated on an SEA knee-joint test bench, and experimental results are compared with simulations to evaluate sim-to-real fidelity.

This work makes three key contributions: (1) the development of a broadly applicable co-design methodology for SEAs, (2) the creation of predictive regression models for the mechanical properties of FDM-printed PLA beams, and (3) the integration of these results into a unified co-design strategy enabling SEAs that leverage additively manufactured compliant elements.