Study Reference Number: 12-5201
Type of activity: Standard study (25 k€)
1 Background and motivation
Legged locomotion is an efficient solution for mobility on unknown and possibly
rugged planetary terrains. Research on walking robots has seen substantial progress in
the last few decades (cf. [1-6]). Studies already focus on the application of walking
robots in space, e.g., . The authors of  even draw inspiration from the jumping
movements of the astronauts that landed on the moon in order to make an initial
design of a jumping robot taking advantage of the moon's low-gravity environment.
While the studies mentioned above (and many more that are omitted here) are
pioneering, a general approach to the design and selection of appropriate walking
gaits for legged robots is lacking. A possible starting point could be to automate the
gait design using evolutionary algorithms [9-11] or other methods of reinforcement
learning  (see Figure 1). Such automatic, rather unbiased, designs of gaits would
allow the solutions to be mostly determined by the problem structure (including the
environment, body and controller). The optimised gaits could then be analysed by
means of the classification methods popular in biological research, leading to generic
insights into the relation between the environmental conditions and the emerged gaits.
In this study, we are mostly interested in the relation between walking gaits and
environmental conditions such as gravity level and soil type.
: Above: Evolved morphologies under different task descriptions, from .
Below: Robot used for evolving walking gaits in .
Milton Hildebrand pioneered the scientific analysis and classification of gaits .
Following his work the movement of each limb is partitioned into a stance phase, and
a swing phase. In the first, the foot is in contact with the ground, while in the latter the
foot is lifted and moved. Each leg completes a cycle in the same length of time in
order for a steady pattern to occur. Thus, any gait can be described in terms of the
beginning and end of stance phase of each limb with respect to each other (see Figure
: diagrams for analysing different gaits, as introduced by Hildebrand .
Early work on six-legged robots used the Hildebrand methodology to analyse the
walking gaits [14,15]. For example, in , this scheme has been used to analyse the
evolved walking gait of a hexapod.
More complex and detailed ways of analysing gaits than the Hidebrand diagrams are
used today, both in biology and robotics. For example, newly studied features include
the contact forces (cf. ). The use of such more complex analysis methods is
encouraged in the context of this project, as long as the information on the gait present
in the Hildebrand diagrams is preserved. Namely, that information captures gait
properties that are essential to the project, such as whether the robot is walking or
2 Study objective
The goal of the Ariadna project is to systematically use Hildebrand diagrams or some
equivalent methodology to analyse walking gaits emerging from an optimisation
process in order to verify whether behavioral switches / bifurcations exist with respect
to parameters such as gravity and soil type.
3 Proposed Methodology
The following methodology is proposed for this study, and should be discussed in the
proposal, though argued alternatives are welcome as long as they promise to achieve
the project goals.
1) A set of experiments tailored to gathering data on optimised walking gaits for
different gravity / soil conditions.The experimental set-up (dynamics, body
morphology, soil) has to be proposed having in mind its relevance to space
exploration. The range of gravities studied should cover different scenarios, ranging
from asteroid- to Jupiter-like environments and beyond. Of course, the relation
between robot mass and gravity is an essential aspect that needs to be discussed.With
respect to the set-up, the proposal will have to address the important issue of the
trade-off between the simulator's speed and realism. Clearly, to reach the project objective many evolutionary runs have to be performed, requiring a fast simulator.
However, the dynamical model implemented also needs a certain level of realism, so
that the optimized gaits are plausible. For example, this may have implications for the
modelling of the soil and contact with the soil.Universities are encouraged to address
this in the proposal, for example by discussing the complexity of simulating specific
soils versus coarse soil categories (e.g., dust, sand, silt, clay, solid). The terrain type
could also be of interest (e.g., flat, undulating, rocky).
The proposed set-up will also have to include an objective function for the
optimization process. While in many robotic studies only speed is optimized, it is
important to keep in mind that in space applications energy efficiency is a critical
parameter as well. The emergence of behavioral switches and bifurcations, and the
critical values of gravity and soil parameters these will occur at, will depend on the
selected objective function.
2) An automated classification scheme to classify gaits, for instance based on the
Hildebrand methodology. The classifications may relate to classes as recognized in
the biological literature. However, methods can also be proposed that automatically
recognize relevant regularities in the different evolved gaits, for instance as in .
Another option would be to perform unsupervised clustering of the gaits.
3) The application of the proposed scheme to the evolved gaits as to locate
switches and bifurcation with respect to a gravity parameter, and other
parameters describing soil properties.
The study can be performed on fixed morphologies (bodies) as well as on
morphologies that are, themselves, the result of the optimisation procedure (as in
). If the latter option is chosen, a degree of realism has to be ensured on the
employed morphologies so that the simulated robots are physically plausible. In
addition, in such a case, the proposal has to specify how the effects of the
environmental conditions on the evolved morphology and on the control can be
This Ariadna project proposal is addressed at research groups with expertise in
any of the following domains:
walking robots, animats, artificial life, evolutionary
robotics, biologically inspired robotics, biomimetics
4 ACT Contribution
Researchers in the ACT will provide space knowledge and perform gait optimisation
in parallel with the proposing group (using the same common simulation platform
agreed). This will result in larger quantities of data and also form a test of the
reproducibility of the methodology.
The analysis of the evolved gaits will also be set up (and performed) in collaboration
with the ACT researchers. The project PaGMO / PyGMO could be used for the
optimisation part of the project, in which case the ACT will also provide support on
the possible software integration with the simulator.
 How to Keep From Falling Forward: Elementary Swing Leg Action for Passive
Dynamic Walkers, by M. Wisse, A.L. Schwab, R.Q. van der Linde, and F.C.T. van
der Helm, IEEE Transactions on Robotics 21 (3), pp. 393-401, (2005)
 Footstep Planning for the Honda ASIMO Humanoid, Chestnutt, J., Lau, M.,
Cheung, G., Kuffner, J., Hodgins, J., and Kanade, T., in ICRA 2005, pp. 629 - 634
 A Bipedal Walking Robot with Efficient and Human-Like Gait. Collins, S.H.,
Ruina, A., ICRA 2005, pp. 1983 - 1988 (2005)
 Online trajectory generation for omnidirectional biped walking, by Behnke, S., in
ICRA 2006. pp. 1597 - 1603, (2006).
 Human-like walking with knee stretched, heel-contact and toe-off motion by a
humanoid robot, by Ogura, Y., Shimomura, K., Kondo, A., Morishima, A., Okubo, T.,
Momoki, S., Hun-ok Lim, and Takanishi, A., in 2006 IEEE/RSJ International
Conference on Intelligent Robots and Systems, pp. 3976 - 3981 (2006).
 A brief review of bipedal robotics research, by Mattias Wahde , Jimmy Pettersson,
in Proceedings of the 8th Mechatronics Forum International Conference, (2002).
 ATHLETE: a cargo handling and manipulation robot for the moon, by B.H.
Wilcox, T. Litwin, J. Biesiadecki, J. Matthews, M. Heverly, and J. Morrison. Journal
of Field Robotics 24 (5), pp. 421-434, (2007).
 Simulation study of a bipedal robot jumping motion approach on moon gravity, by
A.M.M. Omer, H. Lim, and A. Takanishi. IEEE ROBIO 2010, Tianjin, China. (2010)
 Evolving legged robots using biologically inspired optimization strategies, B.
Smith, C.M. Saaj, and E. Allouis. IEEE ROBIO 2010, Tianjin, China. (2010)
 Evolving robot gaits in hardware: the HyperNEAT generative encoding vs.
parameter optimization, J. Yosinski et al. In Proceedings of the European Conference on Artificial Life (ECAL 2011).
 Hornby, Gregory S., Lipson, Hod, and Pollack, Jordan B. Generative
Representations for the Automated Design of Modular Physical Robots. IEEE
Transactions on Robotics and Automation, vol. 19, no. 3, pp. 703-719 (2003).
 The Design of LEO: a 2D Bipedal Walking Robot for Online Autonomous
Reinforcement Learning, by E. Schuitema, M. Wisse, T. Ramakers, and P. Jonker, in
IEEE/RSJ International Conference on Intelligent Robots and Systems (2010)
 The Quadrupedal Gaits of Vertebrates, by Milton Hildebrand, BioScience, Vol.
39, No. 11, Animals in Motion (Dec., 1989), pp. 766-775
 A Distributed Neural Network Architecture for Hexapod Robot Locomotion, by
R.D. Beer, H.J. Chiel, R.D. Quinn, K.S. Espenschied, and P. Larsson, in Neural
Computation 1992 4:3, pp. 356-365
 Robust Agent Control of an Autonomous Robot with Many Sensors and
Actuators, by C. Ferrell, 1993, Technical report AITR-1443, MIT, USA.
 A machine learning approach for automated recognition of movement patterns
using basic, kinetic and kinematic gait data, by R. Begg, J. Kamruzzaman, in Journal
of Biomechanics 38, 2005, pp. 401–408