Learning aerodynamics for the control of flying humanoid robots

Area of Science:

• Robotics
• Aerospace Engineering
• Control Systems

Background:

• Multi-modal locomotion enhances robot versatility in diverse environments.
• Humanoid robots can achieve aerial capabilities through additional actuation.
• Modeling and controlling aerodynamic forces are key challenges for flying humanoid robots.

Purpose of the Study:

• To address the technological and scientific challenges of flying humanoid robots.
• To present the mechanical design and experimental setup for the iRonCub-Mk1 jet-powered humanoid robot.
• To develop and validate models for controlling aerodynamic forces in humanoid robots.

Main Methods:

• Mechanical design and hardware modifications for the iRonCub-Mk1 robot.
• Wind tunnel experiments for precise aerodynamic force and surface pressure measurements.
• Computational Fluid Dynamics (CFD) simulations and an automated framework for aerodynamic dataset expansion.
• Training Deep Neural Network and linear regression models for aerodynamic force control.
• Integration of models into a simulator for designing and validating aerodynamic-aware controllers.

Main Results:

• Successful mechanical design and hardware integration of the iRonCub-Mk1.
• Validation of CFD simulations with wind tunnel experimental data.
• Development of predictive models for aerodynamic forces using machine learning and classical techniques.
• Demonstration of aerodynamic-aware controller design and validation through simulations and physical robot experiments.

Conclusions:

• The iRonCub-Mk1 provides a platform for studying jet-powered humanoid flight.
• Comprehensive aerodynamic modeling and control strategies are feasible for flying humanoid robots.
• The developed models and control approaches enhance the potential for advanced humanoid robot locomotion.