![](bg9.png)
requires only a single, binary signal: if the antenna
or tactile sensor is in contact with an object or not.
Yet, the model captures essential behaviours observed
in its natural paragon, the stick insect antenna (Dürr
et al., 2001). For example, it implements the contact-
induced switch from a large amplitude searching be-
haviour to a local, higher frequency sampling of an
object (Krause et al., 2013b). Contour-Net can be eas-
ily extended to use two or more antennae including
mutual coordination and attentive visual target track-
ing, as will be shown in a follow-up paper.
Raw, minimally pre-processed (normalisation)
data collected from contact events was sufficient to
achieve rotation-, size- and position- invariant shape
recognition rates of over 90% using a plain feed-
forward neural network. A drawback for robotic ap-
plications is that the training dataset needs to be fairly
large. Instead of using raw data as the network input,
extracting higher level features from collected con-
tour points should significantly reduce the required
amount of training samples. Possible features might
be the average Euclidean distance and the spread of
Hopf Oscillator phase differences between successive
contact points. An unsupervised learning algorithm
(SOM; Boltzmann Machine) will be closer to nat-
ural, continuous learning (automatic clustering with
delayed categorization).
We have further shown that incorporating surface
normal information will improve not only the regu-
larity of the sampling trajectory, but also improve the
recognition performance of a neural network. Hit-
ting a surface perpendicular can potentially reduce
slip of an insect antenna inspired robotic tactile sensor
(Kaneko et al., 1995). The reliable and fast detection
of the surface normal using a tactile sensor will be a
challenging and interesting task.
ACKNOWLEDGEMENTS
This work was supported by EU grant EMICAB
(FP7-ICT, grant no. 270182) to Prof. Volker Dürr.
We thank Prof. Holk Cruse for valuable comments on
earlier versions of the manuscript.
REFERENCES
Beyer, K., Goldstein, J., Ramakrishnan, R., and Shaft, U.
(1999). When is "nearest neighbor" meaningful? In
Beeri, C. and Buneman, P., editors, Database Theory -
ICDT 99, volume 1540 of Lecture Notes in Computer
Science, pages 217–235. Springer Berlin Heidelberg.
Brambilla, G., Buchli, J., and Ijspeert, A. (2006). Adap-
tive four legged locomotion control based on nonlin-
ear dynamical systems. In From Animals to Animats
9. Proceedings of the Nineth International Conference
on the Simulation of Adaptive Behavior (SAB 06).
Buchli, J., Righetti, L., and Ijspeert, A. J. (2005). A dy-
namical systems approach to learning: a frequency-
adaptive hopper robot. In Proceedings of the VIIIth
European Conference on Artificial Life (ECAL 2005),
Lecture Notes in Artificial Intelligence, pages 210–
212.
Dürr, V., König, Y., and Kittmann, R. (2001). The anten-
nal motor system of the stick insect Carausius mo-
rosus: Anatomy and antennal movement pattern dur-
ing walking. Journal of Comparative Physiology A,
187(2):131–144.
Erber, J., Kierzek, S., Sander, E., and Grandy, K. (1998).
Tactile learning in the honeybee. Journal of Compar-
ative Physiology A, 183:737–744.
Esslen, J. and Kaissling, K.-E. (1976). Zahl und
verteilung antennaler sensillen bei der honigbiene
(Apis mellifera l.). Zoomorphology, 83:227–251.
10.1007/BF00993511.
Hochreiter, S., Bengio, Y., Frasconi, P., and Schmidhuber, J.
(2001). Gradient flow in recurrent nets: the difficulty
of learning long-term dependencies. In S. C. Kremer,
J. F. K., editor, A Field Guide to Dynamical Recurrent
Neural Networks. IEEE Press.
Hoinville, T., Wehner, R., and Cruse, H. (2012). Learn-
ing and retrieval of memory elements in a navigation
task. In Prescott, T., Lepora, N., Mura, A., and Ver-
schure, P., editors, Biomimetic and Biohybrid Systems,
volume 7375 of Lecture Notes in Computer Science,
pages 120–131. Springer Berlin Heidelberg.
Huang, G.-B., Zhu, Q.-Y., and Siew, C.-K. (2006). Extreme
learning machine: Theory and applications. Neuro-
computing, 70(1-3):489 – 501.
Ijspeert, A. J. (2008). Central pattern generators for loco-
motion control in animals and robots: A review. Neu-
ral Networks, 21(4):642 – 653.
Ijspeert, A. J., Crespi, A., Ryczko, D., and Cabelguen, J.-
M. (2007). From swimming to walking with a sala-
mander robot driven by a spinal cord model. Science,
315(5817):1416–1420.
Kaneko, M., Kanayama, N., and Tsuji, T. (1995). 3-d active
antenna for contact sensing. In IEEE International
Conference on Robotics and Automation, volume 1,
pages 1113–1119. IEEE.
Kaneko, M., Kanayma, N., and Tsuji, T. (1998). Active
antenna for contact sensing. IEEE Transactions on
Robotics and Automation, 14(2):278–291.
Kevan, P. G. and Lane, M. A. (1985). Flower petal micro-
texture is a tactile cue for bees. Proceedings of the
National Academy of Sciences, 82(14):4750–4752.
Kim, D. and Möller, R. (2007). Biomimetic whiskers for
shape recognition. Robotics and Autonomous Systems,
55(3):229–243.
Krause, A. F. and Dürr, V. (2004). Tactile efficiency of in-
sect antennae with two hinge joints. Biological Cy-
bernetics, 91(3):168–181.
Krause, A. F. and Dürr, V. (2012). Active tactile sampling
ICAART2014-InternationalConferenceonAgentsandArtificialIntelligence
100