Corresponding author: Matej Hoffmann
E-mail: matej.hoffmann@iit.it
Journal of Bionic Engineering 14 (2017) 1–14
The Merits of Passive Compliant Joints in Legged Locomotion: Fast Learning,
Superior Energy Efficiency and Versatile Sensing in a Quadruped Robot
Matej Hoffmann
1,2
, Jakub Simanek
3
1. iCub Facility, Istituto Italiano di Tecnologia, Genova 16123, Italy
2. Artificial Intelligence Laboratory, Department of Informatics, University of Zurich, Zurich 8050, Switzerland
3. Department of Measurement, Faculty of Electrical Engineering, Czech Technical University in Prague, Prague 166 27, Czech Republic
Abstract
A quadruped robot with four actuated hip joints and four passive highly compliant knee joints is used to demonstrate the
potential of underactuation from two standpoints: learning locomotion and perception. First, we show that: (i) forward loco-
motion on flat ground can be learned rapidly (minutes of optimization time); (ii) a simulation study reveals that a passive knee
configuration leads to faster, more stable, and more efficient locomotion than a variant of the robot with active knees; (iii) the
robot is capable of learning turning gaits as well. The merits of underactuation (reduced controller complexity, weight, and
energy consumption) are thus preserved without compromising the versatility of behavior. Direct optimization on the reduced
space of active joints leads to effective learning of model-free controllers. Second, we find passive compliant joints with po-
tentiometers to effectively complement inertial sensors in a velocity estimation task and to outperform inertial and pressure
sensors in a terrain detection task. Encoders on passive compliant joints thus constitute a cheap and compact but powerful
sensing device that gauges joint position and force/torque, and — if mounted more distally than the last actuated joints in a
legged robot — it delivers valuable information about the interaction of the robot with the ground.
Keywords: legged robots, underactuated robots, compliant joints, learning locomotion, force and tactile sensing, haptic terrain
classification
Copyright © 2017, Jilin University. Published by Elsevier Limited and Science Press. All rights reserved.
doi: 10.1016/S1672-6529(16)60374-8
1 Introduction
Typical robot designs feature independent actuation
of every Degree of Freedom (DoF), with emphasis on
speed and accuracy of movements, which is best
achieved if the stiffness of all the components is high. In
this domain, the performance of robots has already
surpassed that of humans and animals, as demonstrated
by the industrial robots in assembly lines, for example.
However, these are precisely controlled environments
that facilitate the robots’ tasks. In unstructured envi-
ronments, machines are notably lagging behind animals.
The same strategy that is applied in manipulators —
precise modeling and centralized control of robot and
interaction with environment — faces numerous diffi-
culties: contacts with the environment, which are not
confined to the end-effector anymore, are hard to sense
and model, and the forces that arise cannot be absorbed
by a stiff structure in the long run. Animals, on the other
hand, do not seem to execute trajectories prescribed by a
central controller; instead, behavior is an outcome of the
interaction of several components: the mechanical
properties of the animals’ bodies interacting with the
environment, low-level decentralized spinal control, and,
finally, the signals descending from different brain areas.
During rapid locomotion, the delays in signal transmis-
sion preclude direct feedback control exercised by the
brain
[1]
— the main part thus needs to be offloaded to the
properties of the muscle-tendon system: compliance,
damping, and nonlinear response characteristics in gen-
eral facilitate successful locomotion and even rejection
of disturbances. Mechanical feedback loops, or self-
stabilization
[2–4]
, play a key part and make fast and stable,
yet not centrally controlled, behaviors possible. The fact
that no trajectories are prescribed and enforced, but
negotiated between all the interacting components, is