Project context

Locomotion is a key aspect of our everyday life. From traveling to working, it is at the heart of many of our activities. Yet, mobile robots are usually restricted to operate in environments adapted to their mobility. In a world designed for humans, using humanoid robots is valuable because their body, roughly similar to ours, is potentially perfectly adapted to environments designed for our needs, like ladders or stairs.

There are already many humanoid robots capable of walking. However, they are still far from reaching the impressive human locomotion capabilities, especially regarding energy-efficiency, robustness and behavior richness. Therefore, my work mainly aims at taking inspiration from human locomotion to design bio-inspired algorithms for humanoid robots.

To achieve this purpose, I develop neuromuscular controllers recruiting virtual muscles to drive the robot motors. Because the human neural circuitry commanding these muscles is not known, I make hypotheses about this control scheme to simplify it and to progressively refine the corresponding rules. This project thus offers to improve biped robots locomotion, but also to investigate real human walking, which is far from being understood.

My research is funded by the F.R.S.-FNRS - Fonds de la Recherche Scientifique and by the Walk-Man EU project. In this EU project, two humanoid robots are primarily used as embodiment: the COMAN platform, (Compliant Humanoid) and the WALK-MAN robot. In my work, I mainly use the COMAN platform.

Some parts of my work are presented in the next sections. More information about my work can be found on the Publications page.

COMAN robot

WALK-MAN robot

Gaits generation

The neural circuitry commanding the virtual muscles recruits unknown parameters which must be tuned to obtain the desired behavior (reaching a target speed, minimizing metabolic energy consumption...). Because this tuning is complex and dangerous to perform on the real robot, a simulation environment is first built using the Robotran simulator, with the purpose to reduce the reality gap.

This tuning is then performed in simulation using a Particle Swarm Optimization algorithm. Different particles explore the search space over successive generations to progressively refine the gait.

Real hardware experiment

Once robust gaits are obtained in simulation, the optimized controllers are ported to the real platform without further parameters modification. In the next video, a 2D walking controller is tested. In other words, no lateral balance is implemented. Consequently, I provide lateral stability to the robot while letting it free to move (and possibly to fall) in the other planes.

Gait modulation

In parallel to the hardware experiments, the gait controllers are incremented to add new features. In particular, locomotion richness is obtained by modulating a few controller parameters to steer the gait, resulting in speed and heading adaptation. In the video below, the step length and frequency are modulated to control the walker speed in 2D simulation scenarios. The extension of this modulation for speed and heading control in 3D environments will be added once the related papers will be accepted.

Media

During my Ph.D., I was interviewed by several media (newspaper and TV) about my research.