AI-Enabled Control System Steers Autonomous Drones Through Uncertain Skies
Navigating the skies is a complex task, especially for autonomous drones facing unpredictable weather. Researchers at MIT have developed an innovative AI-enabled control system designed to help drones maintain their course even in turbulent conditions. This new system could revolutionize how drones are used for critical tasks, from fighting wildfires to delivering essential packages.
The challenge lies in the fact that autonomous drones often encounter unforeseen disturbances, such as strong winds. These forces can easily push a drone off its intended trajectory, making it difficult to stay on target. To address this, the MIT team created a machine learning-based adaptive control algorithm capable of minimizing deviations caused by these unpredictable forces.
Unlike traditional methods that require pre-programmed knowledge of potential disturbances, this new technique learns from observational data gathered during a mere 15 minutes of flight time. The AI model adapts in real-time, without needing prior information about the specific nature of the disruptions.
A key feature of this system is its ability to automatically select the most suitable optimization algorithm to adapt to different types of disturbances. By using a technique called meta-learning, the control system learns how to adapt to various scenarios simultaneously, enhancing its tracking performance.
Simulations have shown that this adaptive control system reduces trajectory tracking error by 50% compared to baseline methods. It also demonstrates superior performance in handling new wind speeds not encountered during training. This suggests that the system is highly effective in real-world conditions.
According to Navid Azizan, the Esther and Harold E. Edgerton Assistant Professor at MIT, “The concurrent learning of these components is what gives our method its strength. By leveraging meta-learning, our controller can automatically make choices that will be best for quick adaptation.”
The researchers replaced the function that contains some structure of potential disturbances with a neural network model that learns to approximate them from data. In this way, they don’t need to have an a priori structure of the wind speeds this drone could encounter in advance.
The method also uses an algorithm to automatically select the right mirror-descent function while learning the neural network model from data, rather than assuming a user has the ideal function picked out already. The researchers give this algorithm a range of functions to pick from, and it finds the one that best fits the problem at hand.
“Choosing a good distance-generating function to construct the right mirror-descent adaptation matters a lot in getting the right algorithm to reduce the tracking error,” Tang adds.
The potential applications of this technology are vast. Imagine drones efficiently delivering packages despite strong winds or monitoring fire-prone areas in national parks with greater precision. The new control system could pave the way for more reliable and versatile autonomous drones.
The research team is now conducting hardware experiments to further test their control system on real drones under various wind conditions. They also aim to expand the system’s capabilities to handle multiple disturbances simultaneously and explore continual learning, enabling drones to adapt to new challenges without retraining on past data.
