Introduction
Aerial Manipulators (AMs)—unmanned aerial vehicles equipped with robotic arms—offer significant potential for complex interaction tasks such as infrastructure repair and remote transport. However, the dynamic coupling between the flight platform and the manipulator makes control inherently challenging. This complexity scales significantly when multiple AMs must collaboratively grasp and transport an object. Traditional centralized control architectures, while effective for small groups, suffer from high communication overhead, high computational costs, and a lack of scalability. This paper proposes a decentralized control scheme based on passivity theory to enable stable, modular, and scalable collaborative grasping without the need for a centralized coordinator.
Methodology
Dynamic Decoupling via Coordinate Transformation
The core challenge in AM control is the bidirectional influence between the drone's flight dynamics and the arm's movement. The authors address this through Inertial Decoupling via Coordinate Transformation. By applying a specific mathematical transformation to the equations of motion, the complex dynamics are simplified into a form that resembles a standard robotic manipulator. This transformation effectively "uncouples" the system, making it more amenable to standard control techniques.
Passivity-Based Decentralized Control
The control architecture is built on the principles of Passivity-Based Control (PBC). In control theory, a passive system is one that does not produce its own energy. The authors demonstrate that the interaction between the drones and the grasped object can be modeled as a feedback interconnection of passive systems. A fundamental property of such systems is that their interconnection remains passive, which guarantees that the entire multi-agent group will remain stable regardless of how many drones are added to the task.
Each drone operates using its own local Decentralized Impedance Controller. This controller maintains a specific relationship between force and motion at the gripper, allowing each drone to "feel" and respond to the object and other drones as part of the environment. For flight stability, a geometric tracking controller is utilized to manage the quadrotor's orientation and thrust.
Key Insights and Theoretical Contributions
- Automatic Convergence: A significant finding is that the drones do not require explicit instructions on where to hover to balance a load. Due to the passivity property, the physical interaction naturally pulls the drones into a configuration where they balance each other's forces.
- Modularity and Scalability: Because each drone's control law depends only on its internal state and the forces at its gripper, the system is truly modular. Drones can be added or removed from a task without needing to recalculate a global control strategy.
- Robustness to Sensor Limitations: While the ideal controller utilizes Force/Torque (F/T) sensors at the gripper, the authors performed an Input-to-State Stability (ISS) analysis. This proves that the system can maintain stability even if these sensors are omitted, though with a slight reduction in precision.
Experimental Validation
The proposed control laws were validated using the PyBullet physics engine, simulating quadrotors equipped with 3-DOF planar manipulators. The researchers conducted two primary simulation scenarios:
- 2-AM Collaborative Lift: Two drones successfully approached, grasped, and lifted a 1kg object. The simulation demonstrated the drones tilting appropriately to generate the necessary horizontal grasping forces.
- 10-AM Scalability Test: To prove the decentralized nature of the system, 10 drones were simulated lifting a 5kg object. The group functioned as smoothly as the pair, confirming that the computational and communication requirements do not scale with the number of agents.
Limitations and Future Work
While the simulation results are promising, the authors identify several areas for further development:
- Real-World Factors: The current study lacks experimental data on physical hardware. Future work must account for aerodynamic turbulence (downwash) and communication latency.
- Hardware Constraints: The requirement for F/T sensors at the end-effector may be prohibitive for smaller, lightweight aerial platforms.
- Geometric Assumptions: The model assumes the Center of Mass (CoM) is near the geometric center of the drone. Future research will explore more complex arm geometries and redundant manipulators.
Summary updated