Wednesday, November 11, 2026
Research

TU Delft develops algorithm that enables drones to cooperate when carrying payloads

Transporting heavy payloads with drones remains a major challenge today. A single multirotor can only lift a limited amount of weight and quickly becomes unstable in strong winds or complex environments such as offshore areas or mountainous terrain. Researchers at TU Delft have now developed a solution that enables multiple autonomous drones to work together. A smart control algorithm allows them to lift, position, and stabilize heavy objects—even when the payload starts to swing. The results of the study have been published in Science Robotics.

Many drones make light work

“A single drone can lift only a very small amount of weight,” explains robotics researcher Sihao Sun. As a result, drones are currently of limited use for tasks such as delivering construction materials, transporting crops from remote fields, or assisting in rescue operations.

To address this, the research team designed a system in which multiple drones are attached to the same payload via cables. By constantly adjusting their position and thrust in coordination, the drones can not only share the lifting load but also control the orientation of the payload. This is crucial when something needs to be placed with great precision in a complex or cluttered environment.

Fast reaction is key

The real challenge lies in cooperation. Once drones are physically connected to each other through the payload, every movement affects all aircraft simultaneously. Traditional control algorithms are too slow and rigid to maintain stability under such conditions.

According to Sun, the new algorithm is designed to respond rapidly to disturbances such as gusts of wind or sudden shifts in the payload. The system remains smooth and stable without the need for additional sensors on the object itself—an important advantage for real-world use.

Testing with obstacles, wind, and a basketball

For testing, the researchers built their own multicopters and ran a series of realistic experiments. Up to four drones were used simultaneously, while fans simulated wind and a basketball served as a dynamic payload capable of moving unpredictably. The drones operated fully autonomously, avoiding obstacles and constantly compensating for any motion that could upset their balance. “You simply tell them where to go, and they do the rest,” says Sun.

At this stage, the system still relies on external cameras for positioning and is therefore limited to indoor environments. The next step is to make the technology robust enough for outdoor use, where drones can perceive and navigate their surroundings independently. Potential applications include construction projects, agricultural logistics, and emergency response in areas that are difficult to access.

How it differs from the Australian approach

TU Delft is not the only institution exploring cooperative heavy-lift drones. In Australia, researchers recently demonstrated a system in which two cargodrones can jointly carry payloads of up to 100 kilograms.

Sun notes that the Delft approach distinguishes itself by offering far greater control and flexibility. The system not only lifts but also stabilizes and rotates the payload—a key capability for precision tasks such as assembly. It can also scale up to multiple drones, as demonstrated in lab experiments with four drones and simulations with as many as nine.

Furthermore, the TU Delft algorithm takes into account the individual thrust limits of each drone, preventing overloads and enhancing safety. According to Sun, the underlying control method uses real-time trajectory optimization — “a technique similar to that employed by SpaceX’s Starship during its precision landings.”

Wiebe de Jager

Wiebe de Jager is the founder of Dronewatch (available in Dutch and English). Wiebe is an experienced drone pilot (EASA Specific category certified) and has published a number of bestselling books about drone photography and cinematography.

Leave a Reply

Sign up for our Newsletter