Development of an Intrinsically Safe Actuation System with Adaptive Neuromorphic Control for Humanoid Robotics Application

  • Folgheraiter, Michele (PI)
  • Gini, Giuseppina (Other Faculty/Researcher)
  • Rubagotti, Matteo Rubagotti IT (Other Faculty/Researcher)

Project: Monitored by Research Administration

Project Details

Grant Program

Faculty Development Competitive Research Grant Program 2021-2023

Project Description

This project aims at formalizing, implementing, and testing an intrinsically safe actuation system endowed with adaptive control capabilities. The actuation system is intended to be integrated in the locomotion and manipulation structures of a humanoid robot required to safely interact with its environment. As main design guidelines we seek lightweight, inherently safety, low energy consumption, low cost and adaptability. The outcomes of this project will be both scientific and technological in nature. On the one hand we want to advance the state-of-the-art of self-adaptive control systems that will allow the robot to optimize its operation online while dynamically responding to changes in its environment. On the other hand, we aim at developing an actuation system technology more suitable for robots meant to interact with humans. This research will continue the work started with a previously founded project targeted at realizing and controlling a lightweight biped robot for application in the household and public environment. The actuation system we intend to develop will be based on the combination of a compact high-power and high-speed brushless DC motor equipped with a planetary gearbox and a tunable elastic element. As a novelty, with respect other state-of-the-art actuation systems that integrate fixed elastic components, the actuator we want to develop will be based on a pneumatic artificial muscle which stiffness and damping can be adapted by controlling the inner pressure. The elastic element will be connected in parallel to the output shaft of the motor's gearbox by means of a tendon. This will eliminate the backlash of the gearbox, lower the output impedance of the joint, and act as a storage energy system to reduce the robot energy consumption. By effectively reducing the output impedance of the joints, the elastic elements will increase the robot safety in case of collision with humans or objects present in the environment. Finally, by measuring the elastic elements elongation and knowing their pressure-stiffness characteristic will be possible to estimate the interaction forces and allow implementing a fine force control system.

The presence of non-linear and dynamic viscos-elastic elements in the actuation system of the robot will demand for adaptive control strategies based on a combination of error-feedback and model-based feed forward control actions. As a novelty the feed forward control modules will be based on a Chaotic Recurrent Neural Network CRNN, i.e. a particular category of RNN strongly inspired by the morphology and physiology of biological neural networks. Learning will be possible thanks to an online-capable adaptation module that will be responsible to train the CRNN while the robot is operating.

As a test bed to measure and evaluate the performances of our actuation and control system we will employ the biped robot that we developed in our laboratory. We intend to redesign and realize some of the robot's parts in order to accommodate the new actuation system and implement a completely new torso equipped with arms. The additional Degrees of Freedom (DOFs) will help to stabilize the motion of the robot when executing a gait and allowing us to test the new hardware and control algorithms in a more dynamic condition. Different experiments will be conducted to asses the benefits of introducing tunable elastic elements in the actuation structure and to test the ability of the control system to improve its performance while the robot is operating. This, in terms of rejecting disturbances forces, reducing the robot energy consumption, reducing the impact forces and improving the force control sensibility while interacting with humans.
StatusNot started


  • Humanoid Robotics
  • Bio-inspired Control Systems
  • Recurrent Neural Networks
  • Compliant Actuation Systems


Explore the research topics touched on by this project. These labels are generated based on the underlying awards/grants. Together they form a unique fingerprint.