The new FALCON AI system uses reinforcement learning, which allows it to recognize and adapt to turbulence in real time.
A new pioneer published in NPJ Robotics details the creation of a control system based on artificial intelligence (AI) aimed at reduce the impact of turbulence in dynamic structures, focusing mainly on unmanned aerial vehicles (UAV). This development offers promising potential for smoother UAV flights even in unpredictable atmospheric conditions.
Turbulence, known to cause sudden tremors during changes in air pressure is a challenge for UAVs, which lack the natural adaptability that animals use to navigate such fluctuations. The new study reveals how the FALCON AI system can automatically adjust an aircraft’s controls to compensate for turbulence in real time, writes .
Unlike previous AI systems, typically designed for specific conditions, FALCON utilizes the reinforcement learning to adapt universally. This machine learning technique allows FALCON to understand and react to various turbulence patterns, rather than being limited to pre-defined scenarios.
Central to this approach is Fourier analysis, which uses wave functions to digitally represent wind conditions. This method aligns well with the periodic nature of wind patterns, making FALCON more effective at recognizing and adjusting to turbulence.
Hever Moncayo, professor at Embry-Riddle Aeronautical University, highlighted the potential of the innovation, referring to its compatibility with modern computer systems such as Jetson, which support real-time learningFourier analysis and adaptive control. These systems could integrate seamlessly into FALCON, increasing its effectiveness in managing turbulence.
The research team tested FALCON in the Caltech wind tunnel, equipping a model airfoil with pressure sensors and control mechanisms. Positioned upstream, a moving cylinder introduced random turbulence, simulating real conditions. In just nine minutes, FALCON successfully adapted to maintain airfoil stabilityadjusting pitch and yaw based on immediate pressure readings.
“The Caltech tests demonstrate that FALCON learns in minutes, showing the scalability potential for larger aircraft,” said Moncayo. However, the transition to real-world application will require further investigation, especially in dealing with diverse and unpredictable atmospheric conditions. There also continue to be challenges in validating system performance across various UAV types and configurations.
Additionally, researchers are considering the possibility of data sharing between aircraft to provide early warnings of turbulence, which could significantly improve flight stability. However, such advancements will require rigorous cybersecurity protocols to prevent unauthorized access to critical aircraft data.
