Designing an Energy-efficient Full-sized Humanoid Robot with Fast Adaptive Model-based Neuromorphic Control Architecture.

Faculty Development Competitive Research Grant Program (General) 2024-2026

This project aims at developing a full-sized, lightweight, and low-cost humanoid robot suitable as a test bed to validate fast adaptive neuromorphic control strategies.
Based on the outcomes of a previously founded project, we intend to continue the development of a bipedal robot integrating the upper body and enhancing its structure with compliant servomotors.
The purpose of this research is twofold: on the one hand we aim at advancing bioinspired control strategies targeted at improving the adaptability and flexibility of humanoid robots when executing locomotion and manipulation tasks. On the other hand, we aim to push the boundaries of humanoid robotics technology by developing systems that are lighter in weight and more safer and energy efficient.
The robot's morphology and its kinematic architecture will be designed to partially resemble the human body, enabling it to achieve comparable locomotion and manipulation capabilities. To facilitate safe interaction with humans, the robot's actuation system will incorporate active elastic elements capable of dynamically modulating their stiffness, thereby enhancing the robot's performance, and ensuring safe interaction with its human counterparts.
Exploiting approaches from neuroscience, cybernetics, and artificial intelligence, a hierarchical control architecture will be formulated, implemented, and integrated into the humanoid robot. The central part of this architecture will be a reservoir-based Recurrent Neural Networks (RNN), which resembles the structure of specific microcircuits found in the human prefrontal cortex. This powerful computation paradigm will be used to identify the dynamic model of the robot and keep it up to date while the robot is operating, implement adaptive model-based control strategies and realize pattern generators to control the robot limbs.