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

Abstract

This paper outlines 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 is presented.

The strategic student approach for life-long exploration and learning

“…Artificial development [57] will require particular structures that will guide exploration and learning beyond what can be addressed by pure measures of learning progress. These mechanisms include maturational constraints [5], [6], [58], the development of intrinsic rewards [24], [59], pre-dispositions to detect meaningful salient events, among many other aspects.…”

Sitting pose generation using genetic algorithm for NAO humanoid robots

“…shape and height). However, there are many previous studies on how to control the humanoid robot motion [8], [17], [13], [2], [10]. The work in [8] focuses on asymmetric motion generation for some predefined yoga poses and finds stable trajectory between random initial poses set and a goal pose posture using the evolutionary genetic algorithm approach (GA).…”

“…Based on these keyframes in which express the motion as a combination of a pose and a time frame, genetic algorithm can be applied in order to generate the optimal stable motion. Machine learning has been used in [10] for teaching humanoid robot on biped dynamic walking. This work invests the basic primitives in human behavior and use them as constraints in the learning process in order to generate human-like motion.…”

The poppy humanoid robot: Leg design for biped locomotion

“…On one hand, a way to permit robots to adapt their behaviors to unknown environments is to provide them with control algorithms which can be updated with learning algorithms based on social guidance [6], or on autonomous self-exploration [4] [17]. On the other hand, a part of the computation needed for such adaptation could also be done through the intrinsic mechanics and electronics of the robot, thus providing effective and hyper-responsive reactions while simplifying the algorithms of the different behaviors.…”

“…A central vision of this research program, deeply inspired by infant learning and development, is that life-long skill learning in the real world can only effectively happen if statistical inference is guided by strong constraints, in particular *This research was partially funded by ERC Starting Grant EXPLORERS 240007. 1 INRIA Flowers Team, Bordeaux, France matthieu.lapeyre, pierre.rouanet, pierre-yves.oudeyer at inria.fr related to the physics of the body (their material, their geometry and the evolution of this geometry as the body grows [5][3] [17]) and to the social environment [6] [1]. Indeed, typical humanoid bodies are high-dimensional, which is extremely challenging for acquiring sensorimotor controller.…”