Quantification of Morphological Computation in Legged Locomotion

: farscope thesis

Abstract

Active Control (AC), a widely used approach in rigid-body robotics, employs automated feedback control with sensors and actuators for precise regulation based on well-established
control theories. While highly effective in industrial settings, its application to legged
locomotion remains challenging due to the unpredictable nature of open-world environments,
making accurate modeling difficult. In complete contrast to AC, Passive Dynamic Walkers use
their natural dynamics to move downhill with no actuation or control. Furthermore, biological
systems, operating under several constraints, have evolved an effective control approach, where
they leverage their physical structures to perform functions that might otherwise rely entirely on
their nervous system. This is often referred to as Morphological Computation (MC), and along
with other factors such as control approaches and sensory feedback, MC significantly contributes
to the ability of natural systems to demonstrate robust locomotion in unstructured environments.
Inspired by biology, researchers have developed legged robots that use a blend of AC and MC
using different approaches such as by adding passive elements and several others. Consequently,
MC has played an important role in enhancing locomotion performance and simplifying control.
To take advantage of MC fully and more systematically, quantifying MC in legged robot locomotion
is essential for assessing its impact and enable effective robot modeling and design, ensuring
adaptability to specific applications. The principles of MC have been increasingly studied in recent
years and researchers have investigated theoretical approaches to quantify the contribution of MC
for a variety of robotic systems. However, the majority of these studies have relied primarily on
analytical and/or simulated models. Since MC fundamentally depends on real-world interactions,
experimental validation on physical robotic platforms is crucial. However, current research lacks
comprehensive studies exploring MC in real-world conditions, highlighting a significant gap in
the field. The aim of this research is to quantify MC in a real legged robotic system.
This research is decomposed into three objectives: develop a simulated model of a legged
robot to quantify MC, build a bespoke test rig to study a robotic leg for quantifying MC and to
investigate the benefits of MC with respect to energy consumption and Cost of Transport (COT).
Firstly, after investigating different methods, I developed the algorithm for quantifying MC in
legged robots by adopting the information theoretic method proposed by Ghazi-Zahedi et al. [32].
The core principles behind the method and the algorithm have been described, including how
they can be adapted for a simulated and legged robot. Next, I developed the simulated quadruped
model to quantify MC on a periodic hopping gait and explored the variations in MC over three
different joint stiffness levels (representing a morphological change). The results revealed a clear
relationship between stiffness and MC, where MC increases as stiffness decreases. Furthermore,
this simulation serves as an analytical toolbox to facilitate informed design modifications, and
evaluate changes at the modeling stage before constructing physical robots. Finally, I built a
bespoke test rig with a two degree of freedom (2DOF) leg, by substantively modifying an open
source design. I generated a walking gait and quantified the MC throughout the trajectory.
To explore variations in MC over different morphological changes, I built a compliant elastic
leg segment, inspired by prosthetic limb designs. MC was quantified across four different leg
configurations: the original leg and the elastic leg with three different thicknesses, resulting
in different stiffness levels with the original leg being the highest. The results demonstrated a
clear inverse relationship between stiffness and MC, with MC increasing as stiffness decreased,
confirming the simulation findings. Furthermore, the Cost of Transport (COT) analysis revealed
that the stiffest leg exhibited the highest COT (lowest energy efficiency), while the most compliant
leg achieved the lowest COT (highest energy efficiency).
The results demonstrate how real-world measurements of MC may help design the morphology
for improved locomotion robots. For future work, the test rig can serve as an experimental
platform to quantify MC on any locomotion robot with an arbitrary gait, as already proven in one
of the publications. Different morphological variations such as materials, dampness variations,
environments, etc. can be applied to fine-tune morphological variations.

Date of Award	4 Nov 2025
Original language	English
Awarding Institution	University of Bristol
Supervisor	Andrew T Conn (Supervisor) & Helmut Hauser (Supervisor)