Agile and Cooperative Aerial Manipulation of a Cable-Suspended Load

TL;DR

The paper tackles the challenge of agile aerial manipulation of a cable-suspended load by a team of quadrotors. It introduces a trajectory-based framework that solves a online kinodynamic motion planning problem, generating receding-horizon, dynamically feasible trajectories for all quadrotors while accounting for full load–cable dynamics. An EKF-based load-cable state estimator and an onboard INDI trajectory-tracking controller enable robust real-time operation without sensors on the load, achieving substantial gains in agility (e.g., eightfold accelerations) and resilience to model uncertainties and disturbances. Real-world experiments demonstrate high-speed maneuvers, obstacle avoidance through narrow gaps, wind robustness, and scalability to multiple units, underscoring practical potential for time-critical missions such as search and rescue and precision delivery.

Abstract

Quadrotors can carry slung loads to hard-to-reach locations at high speed. Since a single quadrotor has limited payload capacities, using a team of quadrotors to collaboratively manipulate a heavy object is a scalable and promising solution. However, existing control algorithms for multi-lifting systems only enable low-speed and low-acceleration operations due to the complex dynamic coupling between quadrotors and the load, limiting their use in time-critical missions such as search and rescue. In this work, we present a solution to significantly enhance the agility of cable-suspended multi-lifting systems. Unlike traditional cascaded solutions, we introduce a trajectory-based framework that solves the whole-body kinodynamic motion planning problem online, accounting for the dynamic coupling effects and constraints between the quadrotors and the load. The planned trajectory is provided to the quadrotors as a reference in a receding-horizon fashion and is tracked by an onboard controller that observes and compensates for the cable tension. Real-world experiments demonstrate that our framework can achieve at least eight times greater acceleration than state-of-the-art methods to follow agile trajectories. Our method can even perform complex maneuvers such as flying through narrow passages at high speed. Additionally, it exhibits high robustness against load uncertainties and does not require adding any sensors to the load, demonstrating strong practicality.

Figures (13)

• Figure 1: Snapshot of the real-world experiments. We propose an approach to control a cable-suspended load using multiple quadrotors with high agility. (A) Our approach enables agile full-pose control of a cable-suspended load. (B-D) It enables the quadrotors to dynamically control the load pose and fly through a narrow passage and a horizontally oriented gap. A summary of the experiments is highlighted in Movie 1.
• Figure 2: Performance in tracking the reference Fast. (A) Experiment comparing our method against two baseline methods to follow the reference trajectory Fast, a figure-eight trajectory with a maximum speed of 5 m/s and a maximum acceleration of 8 m/s$^2$. The detailed expression of the reference is given in Table \ref{['tab: algebratic_load_reference']}. (i) Top view of the flight path of the CoG of the load. (ii-iii) Time history of the root-mean-square error of the load position and attitude tracking error of the load. We used axis-angle representation for the attitude error. (B) Experiment comparing our methods with tightened thrust limits and without, while tracking the reference Fast. (i) Top view of the flight path of the CoG of the load. Once the maximum thrust was limited, the reference trajectory became dynamically infeasible for the system to follow accurately (red). (ii-iv) The commanded collective thrust of the three quadrotors with the reduced thrust limits (black dashed lines).
• Figure 3: Obstacle avoidance through dynamic motion. Both tasks were provided with a line segment reference that originally intersected the obstacles. (A) Task 1: Flight through a narrow passage between two walls. (i) Top view of the load center and three quadrotors with predicted trajectories at $t={1.5}~\mathrm{s}$. (ii-iii) Velocity and acceleration profiles. (iv) Distances between quadrotors. (v) Snapshot of the experiment when the multi-lifting system flew through the narrow passage. (vi) Load inclination during traversal, defined as the angle between the load-fixed z-axis and the world-frame z-axis. (B) Task 2: Flight through a horizontally oriented narrow gap. (i) Side view of the trajectory and predicted trajectories at $t={1.5}~\mathrm{s}$. (ii) Snapshot of the experiment when the multi-lifting system flew through the horizontally oriented gap. (iii-iv) Velocity and acceleration profiles. (v) Cable inclinations during traversal, defined as the angle between cable directions and the gravity.
• Figure 4: Test under load model uncertainties and communication delays. (A) Tracking performance of our method versus two baseline methods under various load model mismatches and communication delays while tracking reference Slow, Medium, and Fast as defined in Table \ref{['tab: algebratic_load_reference']}. The baseline methods failed to follow the reference Fast even without mismatches, whereas our method remained robust. Each box corresponds to one run and summarizes the error at 4500 reference points: median (center line), 25th–75th percentiles (box), and whiskers extending to the minimum and maximum non-outlier values; outliers are defined as points lying beyond 1.5 interquartile range (IQR) from the box edges. (i–iii) Position error is in meters. (iv–vi) Attitude error is in degrees, calculated through axis-angle representations. (B) Real-world experiment where a ${0.6}~\mathrm{kg}$ basketball was placed onto the ${1.4}~\mathrm{kg}$ basket-shaped load and introduced a considerable inertia model mismatch of the load. Our method ran without knowing the presence of the basketball. (i) Top view of the path of the load CoG with and without a sloshing load. (ii-iii) Time history of the position and attitude tracking error. (iv) A snapshot of the experiment.
• Figure 5: Test under wind disturbances. (A) Comparison of position error between our approach and the baseline methods at different wind speeds in simulation environments. (B-C) Snapshot of experiments with three and four quadrotors, respectively, under windy conditions generated by a 1.5 m diameter fan. (D-E) Real-world experimental data from three and four quadrotors carrying a load to follow a straight line at a speed of 0.3 m/s in a 5 m/s wind field. (F) Real-world experimental data with four quadrotors carrying the payload and flying over the wind field while following a curved trajectory at a speed of 2 m/s. The videos of the experiments are provided in Movie S4.