Science Robotics: Multi-UAV Cooperative Payload Transport Through a 0.8 m Narrow Gap

How narrow is 0.8 m? When three UAVs cooperatively lift a payload with cables, the overall width of the system in hover configuration is about 1.4 m. Without changing the formation and payload attitude, it cannot pass through a 0.8 m passage at all. The real question is whether the system can combine high-speed maneuvering with stable control in such a narrow gap.

On October 29, 2025, Sihao Sun’s team at Delft University of Technology published a paper in Science Robotics titled “Agile and cooperative aerial manipulation of a cable-suspended load.” The work proposes a centralized cooperative planning and control framework that incorporates obstacle avoidance and collision-avoidance constraints into cooperative decision-making, enabling agile full-pose control of the payload and allowing a multi-UAV suspended-load system to pass through a 0.8 m narrow gap at high speed.

Video source: https://www.youtube.com/watch?v=FBWN-rTK1YU

Technical Challenges

To make multi-UAV suspended-load transport both agile and safe without relying on payload-mounted sensors, the main challenges are:

• Full-pose agile control: payload position and attitude are indirectly controlled through cable tension. During high-speed maneuvers, cable direction and tension change quickly, creating strong coupling, tracking errors, and oscillation.
• Obstacle avoidance and inter-UAV collision avoidance: in narrow environments, the system must avoid obstacles while maintaining safe distances between aircraft. The feasible space becomes compressed, often requiring formation and payload-attitude changes.
• Closed-loop control without payload sensors: without sensors on the payload, the payload state must be estimated from UAV-side information, and estimation errors can be amplified under agile maneuvers.
• Smooth 10 Hz online planning: when trajectories are updated in a receding-horizon manner, new and old predicted trajectories must connect smoothly with high-order continuity and stable initialization.

Research Highlights

The key contribution is the use of online whole-body dynamics motion planning instead of the traditional cascade structure of “outer-loop allocation, inner-loop tracking.” The planner directly generates a feasible cooperative trajectory for the entire system over a future horizon, while onboard robust tracking compensates for payload model mismatch and external disturbances.

As a result, the system no longer focuses on simply distributing forces. Instead, all UAVs cooperate around a single predicted trajectory, enabling high-speed and stable maneuvering for multi-UAV payload transport.

Centralized Cooperative Trajectory Planning

The study uses centralized online planning to generate coordinated trajectories. The planner solves a finite-horizon optimal control problem in a receding-horizon manner, updating at 10 Hz and predicting a feasible trajectory for roughly the next two seconds.

The optimization includes UAV-payload coupling, thrust limits, cable-tension constraints, inter-UAV collision avoidance, and obstacle avoidance, ensuring that the trajectory is both flyable and safe. To avoid trajectory jumps caused by 10 Hz updates, the previous predicted trajectory is resampled to initialize high-order variables related to cable direction and tension, helping the new trajectory connect smoothly.

INDI Robust Tracking

The onboard tracking controller uses incremental nonlinear dynamic inversion (INDI) and IMU feedback to compensate online for disturbances caused by cable forces. Errors in planned cable tension, such as those caused by payload inertia mismatch, can be effectively compensated by the onboard controller, improving robustness to model uncertainty.

Payload Without Sensors

The framework includes a centralized EKF that combines a payload-cable model, UAV state estimates, and IMU measurements to output the payload state and cable directions required by the planner. This reduces the need for payload modification or additional sensor installation.

Experimental Tests

High-Speed Trajectory Tracking

The research team built a three-UAV cooperative transport platform using three modified Agilicious quadrotors carrying a 1.4 kg payload. The tests evaluated whether the system could track trajectories stably at high speed, complete constrained maneuvers safely, and remain reliable under external disturbances and modeling errors.

In figure-eight flight tests, the system tracked trajectories at up to 5 m/s and 8 m/s². Compared with conventional geometric control and NMPC, which became less stable under the same aggressive motion, the proposed system maintained closed-loop controllability and kept payload position error at around 0.2 m.

When thrust limits were tightened further, the system automatically reduced curvature and slowed the motion so that the trajectory remained dynamically feasible, avoiding instability caused by attempting to track an infeasible reference.

Obstacle Tests

Test 1: Passing through a vertical slit with 0.8 m wall spacing. The system automatically adjusted payload attitude so that the load passed through sideways with an attitude angle of about 70 degrees while maintaining safe UAV separation. The passage speed exceeded 4 m/s.

Test 2: Passing through a horizontal gap with 0.6 m height. Instead of slowly squeezing through, the UAV group expanded its formation and rapidly flattened the cables, using momentum to complete the passage. The key crossing motion took about 1.2 seconds.

Wind and Payload Uncertainty

In wind-disturbance experiments with approximately 5 m/s wind, the system still completed flight and trajectory tracking. Errors increased but remained controlled.

For payload uncertainty testing, the researchers added a moving basketball and other additional loads inside the payload, significantly changing the center of mass and inertia. The planning model did not know these changes, yet the system still completed trajectory tracking, showing strong tolerance to payload-model errors.

Outlook

This research is not only relevant to UAV transportation. It may also be extended to other systems that need disturbance suppression and can be especially useful in narrow spaces and complex obstacle environments.

For example, in disaster sites, corridors, or narrow passages, such systems could transport clearing tools and emergency supplies more quickly. In factories and warehouses, they may carry large items through doors and corridors. In construction and maintenance, they may support constrained-space transport and place objects faster, more stably, and more accurately.

Resources

• Paper: Agile and cooperative aerial manipulation of a cable-suspended load
• Journal: Science Robotics, 2025-10-29
• DOI: 10.1126/scirobotics.adu8015
• Project page: https://sihaosun.github.io/ScienceRoboticsCAMLs/
• Paper link: https://www.science.org/stoken/author-tokens/ST-3015/full