TO BE UPDATED…
TO BE UPDATED…
TO BE UPDATED…
HexaQuaBip is a reconfigurable six-, four- and two-legged robot designed to be capable of shaping its morphology and switching between various locomotion modes. The idea is to merge main legged animal morphological factors in a reconfigurable, yet non modular, robot that would offer a great locomotor versatility. Our comparative study of leg segment length proportions reported in biological literature shows that inter-species variability is limited. It suggests which segments of a classical three-segmented leg are more appropriate to bear variable-length mechanisms.
Veinguertener, A., Hoinville, T., Bruneau, O. and Fontaine, J.-G. (2009) Morphological design of the bio-inspired reconfigurable HexaQuaBip robot. In Mobile Robotics: Solutions and Challenges, Proc. of CLAWAR 2009, pp. 205-214.
Veinguertener, A., Hoinville, T., Bruneau, O. and Fontaine, J.-G. (2009) From Morphologies of Six-, Four- and Two-Legged Animals to the HexaQuaBip Robot’s Reconfigurable Kinematics. In Proc. of ICIRA2009, LNAI 5928, pp. 1255-1265.
We proposed parameter constrains applied to plastic neuronal models and inspired by neural homeostasis phenomena, in order to evolve both flexible and stable pattern generators for a single-legged locomotion task (involving conflicting sub-behaviors). The constrained controllers showed better evolvabiliy and behavioral stability than the unconstrained ones. Furthermore, we have found that homeostatic neuronal dynamics can evolve implicitly without relying on any active homeostatic mechanisms and can be implemented through hebbian plasticity, usually considered destabilizing.
Hoinville, T., Tapia Siles, C. and Hénaff, P. (2011) Flexible and Multistable Pattern Generation by Evolving Constrained Plastic Neurocontrollers. Adaptive Behavior, 19(3):187-207.
Hoinville, T. and Hénaff, P. (2004) Comparative study of two homeostatic mechanisms in evolved neural controllers for legged locomotion. In Proc. of IEEE/RSJ IROS 2004, pp. 2624-2629.
Hoinville, T. and Hénaff, P. (2004) Evolving plastic neural controllers stabilized by homeostatic nechanisms for adaptation to a perturbation. In Proc. of ALIFE IX, pp. 81-88.
We investigated evolving plastic CPGs based on neuronal relaxation oscillators subject to neuromodulation. We have shown that extrinsic neuromodulation (ie. the modulatory neurons do not receive any feedback connection from the modulated neurons) led to evolved pattern generators able to adapt the speed and direction of 6-, 8- and 10-legged robots to user commands. Moreover, only a simple fitness function based on the viability notion was necessary.
To achieve this, we used theoretical symmetries inspired from observing animal gaits to constrain the topologies of evolved central pattern generators. We have shown that, combined to a spatial genetic encoding of neuronal connectivity, this modular architecture is an evolvable substrate for multiple legged locomotion gaits.
Hoinville, T. (2007) Évolution de contrôleurs neuronaux plastiques : de la locomotion adaptée vers la locomotion adaptative. PhD Thesis, UVSQ, Versailles, France.
We evolved simple recurrent neural networks to control the balance of a robot skating on two in-line wheels and equipped with two arms. Several evolution runs resulted in gaits stabilizing the robot’s inclination without perturbing too much its direction. Other evolved controllers were found to implement reflex behaviors able to put the robot back up after falling aside.
Hoinville, T., Hénaff, P. and Monacelli, E. (2005) Evolving Simplistic Neural Controllers toward Adaptation to Internal Perturbations. AMSE Journal, Advances in Modelling, Series C, 60(5-6):57-70.
Hoinville, T. (2002) Évolution de contrôleurs neuronaux pour la commande d’un robot à roues et pattes : vers une tolérance aux pannes. MSc Thesis, UPMC Paris 6, France.