Racing Drones Assist Space Agency in Advancing Next-Generation Navigation Technologies

The European Space Agency (ESA) is exploring the dynamic realm of drone racing to test advanced AI-powered control systems, with the goal of enhancing future spacecraft navigation in the unpredictable conditions of space. Unlike Earth, where predictable environments allow for more straightforward navigation, space presents a myriad of challenges, including gravitational fluctuations and atmospheric turbulence, requiring spacecraft to adapt swiftly to maintain their trajectory.

To address this, researchers are developing a cost-effective solution that empowers spacecraft to autonomously adjust their paths without relying solely on manual corrections. This innovative approach, known as Guidance & Control Networks (G&C Nets), enables spacecraft to continuously optimize their trajectories in real time, rather than adhering to a predetermined route.

Dario Izzo, the scientific coordinator for ESA's advanced concepts team, emphasized the advantages of utilizing drone racing as a testing platform for neural networks. “We’ve been investigating the application of trainable neural networks for the autonomous management of complex spacecraft maneuvers, including interplanetary transfers, surface landings, and dockings,” Izzo stated. “By leveraging neural networks, we can enhance onboard operations, thereby increasing mission autonomy and resilience.”

During recent tests at Delft University of Technology’s research facility, affectionately dubbed the “Cyber Zoo,” drones equipped with ESA's AI control systems showcased their abilities in a race against the clock. Within a 33-foot by 33-foot testing area, these drones navigated carefully designed courses, optimizing their speeds while adaptively responding to real-time environmental changes. Although the drones operated independently, they all utilized the same neural network-driven control system, allowing them to continuously recalibrate their optimal paths.

The challenge of bridging the gap between simulated actuator performance and real-world conditions was paramount. Christophe De Wagter, TU Delft’s principal investigator, noted that understanding this "reality gap" was crucial. “When flying, we identify discrepancies between expected and actual performance and teach the neural network how to adapt. For instance, if the propellers produce less thrust than anticipated, the drone detects this with its accelerometers and the neural network adjusts commands to follow the new optimal path,” De Wagter explained.

Insights gained from these drone experiments will be applied in ESA’s upcoming Hera mission, which aims to investigate the aftermath of NASA’s Double Asteroid Redirection Test (DART). In September 2022, NASA launched a projectile towards an asteroid in the Didymos binary system to assess planetary defense strategies against potential threats. Hera will evaluate the impact site of the DART mission while adeptly navigating nearby asteroids and collecting vital data.

ESA envisions future probes and spacecraft that can maneuver around objects like asteroids autonomously and efficiently. “Traditionally, spacecraft maneuvers are meticulously planned on Earth and then executed by the spacecraft,” explained Sebastien Origer, an ESA graduate trainee. “In our proposed end-to-end Guidance & Control Networks framework, all calculations and adjustments occur on the spacecraft itself. Rather than following a fixed path, the spacecraft constantly revises its trajectory based on its current location, resulting in considerably greater efficiency.”

Origer highlighted that the drone-based research “serves as a foundation for building trust, developing a robust theoretical framework, and establishing safety parameters in anticipation of future space mission demonstrations.”