Abstract
This paper addresses the kinematic tracking control problem for aerial manipulators, which often struggle to meet precise time constraints using traditional feedback strategies. We propose a novel framework that combines end-effector tracking control based on a user-defined preset trajectory with quadratic programming (QP)-based reference allocation. This approach ensures the end-effector reaches its target within a specified time while maintaining tracking errors within a predefined performance envelope. Furthermore, the QP-based allocation respects the physical constraints of both the quadcopter base and the robotic arm. Experimental results validate the algorithm's effectiveness in achieving high-precision tasks like aerial grasping and assembly within guaranteed timeframes.
The Challenge of Aerial Manipulation
Aerial manipulation combines the 3D mobility of multirotor drones with the dexterity of robotic arms. However, controlling these systems is complex due to dynamic coupling—where the movement of the arm affects the drone's stability and vice versa—and the inherent unpredictability of 3D flight environments.
Existing kinematic control methods, such as Closed-Loop Inverse Kinematics (CLIK) or standard Model Predictive Control (MPC), face two significant hurdles:
- Lack of Temporal Guarantees: They cannot guarantee exactly when a robot will complete a movement.
- Unconstrained Error Spikes: Tracking errors are not strictly bounded during transit, which can lead to collisions in cluttered environments or failed precision tasks like peg-in-hole assembly.
Proposed Methodology
The researchers developed a two-layer control architecture to address these limitations, integrating trajectory tracking with optimized motion distribution.
1. Preset Trajectory Tracking Control
Instead of a simple point-to-point command, this layer utilizes a Performance Envelope. This mathematical boundary starts wide to accommodate initial positioning errors and shrinks exponentially toward a minimal steady-state value.
By designing a "preset trajectory" for the tracking error, the system ensures the robot reaches its target within a specific time ($t_p$). Crucially, this method avoids the mathematical singularities often found in traditional "Prespecified Performance Control" (PPC) that uses barrier functions, leading to greater stability near the performance boundaries.
2. QP-Based Reference Allocation
Once the desired end-effector velocity is determined, the system must decide how to distribute this motion between the quadcopter base and the high-speed Delta arm. The framework employs Quadratic Programming (QP) to solve this allocation problem. By incorporating physical constraints—such as maximum drone velocity and the joint limits of the Delta arm—directly into the optimization, the controller ensures that the generated commands are always within the hardware's physical capabilities.
Experimental Validation
The system was evaluated using a custom-built aerial manipulator (0.65m quadcopter with a 3-DOF Delta arm) and a Vicon motion capture system for ground-truth data.
Key Results
- Precision: In static tracking tests, the proposed method reduced average tracking error by 22.8% compared to CLIK and 12.0% compared to standard error-based feedback.
- Speed: The system reached its target in a median time of 2.4s, significantly faster than CLIK (4.8s) or error-based (3.2s) approaches.
- Task Success: The robot successfully performed complex tasks, including aerial grasping and a peg-in-hole assembly. In the assembly task, it maintained a terminal tracking error of just 0.89 cm across 16 trials.
Conclusion and Future Directions
This framework provides a robust solution for time-critical aerial manipulation by treating the drone and arm as a single integrated system. Future work will focus on integrating neural networks for better disturbance estimation (e.g., wind) and moving from kinematic control to full dynamic control to enable even more aggressive maneuvers.
Summary updated