Section: New Results

Motor Learning and Morphological Computation

Morphological Computation in Acroban the Humanoid: Balance Control and Dynamic Walking

Participants : Olivier Ly, Pierre-Yves Oudeyer, Matthieu Lapeyre, Jérome Béchu, Paul Fudal, Haylee Fogg.

We have continued to elaborate and experiment the humanoid platform Acroban and its use to study various scientific topics. Our goal was to study three main issues: 1) Compliance and semi-passive dynamics in the framework of dynamic walking in humanoid robots and more generally its impact in terms of semi-passive interactive motor primitives and their robustness to unknown external perturbations; 2) the advantage of a bio-inspired multi-articulated vertebral column in the dynamics of these motor primitives; The platform uses mechatronic components that allow us to adjust dynamically the compliance of actuators, which combines with the intrinsic mechanical compliance of the structure due to the use of elastics and springs. We have explored how these capabilities can allow us to enforce morphological computation in the design of robust dynamic locomotion. Compliance also allows us to design semi-passive motor primitives using the torso as a system of accumulation/release of potential/kinetic energy. This is made possible by the combination of adequate morphology and materials, full-body compliance, semi-passive and self-organized stable dynamics, as well as the possibility to experiment new motor primitives by trial-and-error thanks to light-weightedness. These results were presented in [25] . A dedicated web page with videos is available at: http://flowers.inria.fr/acroban.php .

Maturational constraints for motor learning in high-dimensions: the case of biped walking

Participants : Matthieu Lapeyre, Pierre-Yves Oudeyer, Olivier Ly.

We have elaborated and began to experiment a new developmental approach to motor learning in very high-dimensions, applied to learning biped locomotion in humanoid robots. This approach relies on the formal modeling and coupling of several advanced mechanisms inspired from human development for actively controlling the growth of complexity and harnessing the curse of dimensionality: 1) Maturational constraints for the progressive release of new degrees of freedoms and progressive increase their explorable ranges; 2) Motor synergies; 3) Morphological computation; 4) Social Guidance. An experimental setup involving a simulated version of the Acroban Humanoid robot, based on the V-REP simulator, has been elaborated, and initial encouraging results were obtained. These results are presented in [23] .

Acroban v2: improving morphological computation with dampers

Participant : Olivier Ly.

Theoretical studies and experiments concerning in particular dynamics of passive walkers drove us to design, construct and continue to experiment a new version of Acroban. This new version has two goals both fitting in the study of the impact of morphology in the behaviour of the robot:

• experiment deep structural modifications of the morphology, in order to avoid as much as possible inelastic chocks. Indeed, during the gait, the unstability is mainly due to chock at the landing of the foot.

• improve the global ratio weight/power of the robot in order to get more dynamic movements.

Indeed, this new version uses RX-28 motors which are lighter than the RX-64 motors which are used in the first version of Acroban. The robot is smaller and lighter. First experiments show that the obtained ratio weight/power is better than the first version. Movements of the robot, and in particular amplitude of locomotion movements, are not limited by torque now. Second, we have experimented plastic materials to design the structure in order to makes is naturally flexible comparing to the metal used in the first version. This way, we improve the natural compliance of the robot. Finally, and this is probably the most important change, we used non actuated linear joints in the hip and in the spline. To control these linear joints, instead of servo-motors, we use dampers.This kind of design is new in humanoid robotic. While bringing new control problems (because of the non-controled joints which makes the robot semi-passive), this design softens chocs in a significant maner. Experiment shows that stability of the whole structure is greatly improved especially during locomotion.