Abstract
This research addresses the challenge of maintaining stability and precision during physical interaction between aerial manipulators and environmental surfaces. We present a motion/force controller designed to guarantee the tracking of time-varying trajectories while ensuring stability during the critical transition from free flight to contact. By modeling contact forces using a Kelvin-Voigt linear model and estimating parameters via a recursive least-squares (RLS) estimator, the system adapts to unknown surface properties. The controller employs a disturbance-observer (DOB) and gain-scheduling based on Input-to-State Stability (ISS) conditions. Experimental results on tilted surfaces demonstrate that the proposed framework enables robust tracking and stable transitions even at varying approach speeds.
Introduction
Aerial manipulation—the use of unmanned aerial vehicles (UAVs) equipped with robotic arms—is a burgeoning field with applications in structural inspection, maintenance, and surface processing. However, these systems are inherently underactuated; multirotors must tilt to generate lateral forces, which complicates the application of precise force against a surface.
Existing research often treats free flight and steady-state contact as separate regimes. In practice, the "switching" phase—the moment of impact—is a primary source of instability. Furthermore, many tasks require the drone to follow complex, time-varying force profiles (e.g., varying pressure during cleaning) rather than a constant force. This paper proposes a robust control framework that specifically addresses the stability of the transition phase and the ability to follow dynamic force/motion paths on arbitrarily tilted surfaces.
Methodology
Dynamic Modeling and Decomposition
The authors derive the translational dynamics of the aerial manipulator relative to its end-effector. To manage the underactuated nature of the platform, the dynamics are decomposed into two orthogonal subspaces:
- Force Space: Normal to the contact surface.
- Motion Space: Tangential to the surface.
This decomposition allows the controller to prioritize contact force while simultaneously managing sliding motion along the surface.
Environment Estimation and Robust Control
To account for unknown surface characteristics, the contact force is modeled using the Kelvin-Voigt linear model, representing the surface as a spring-damper system. A Recursive Least-Squares (RLS) estimator identifies the stiffness ($k_e$) and damping ($b_e$) parameters in real-time.
To handle model uncertainties and external disturbances such as wind or friction, a Disturbance Observer (DOB) is integrated into the control loop. This ensures that the nominal model remains accurate even in non-ideal conditions.
Stability-Guaranteed Gain Scheduling
The core innovation of this work is the Force-Controller-Gain Scheduler. The researchers derived analytical Input-to-State Stability (ISS) conditions for the switched system. By solving explicit mathematical inequalities, the system identifies a "stable region" for controller gains ($k_f, b_f$). This approach is computationally efficient, allowing the onboard computer to update gains in real-time to prevent the drone from bouncing or losing contact during the transition from flight to interaction.
Experimental Validation
The framework was validated using a custom-built coaxial octocopter (3.78 kg) equipped with a 2-degree-of-freedom robotic arm and a 1-axis force sensor. Experiments were conducted using an Optitrack motion capture system for ground-truth positioning. The testbed involved a medium-density fibreboard (MDF) surface placed at various tilts.
Key Results
- Time-Varying Tracking: The manipulator successfully tracked complex force trajectories, such as sine waves, while sliding across both vertical and slanted surfaces.
- Impact Robustness: The gain-scheduled controller stabilized contact quickly even at approach speeds of 0.3 m/s, which typically cause significant bounce or instability in fixed-gain systems.
- Adaptive Performance: The RLSE successfully identified environment parameters, allowing the controller to adapt its behavior to the specific material properties of the MDF surface.
Discussion and Future Work
While the proposed controller demonstrates high reliability, it currently assumes the orientation of the surface is known. Future iterations could integrate onboard vision or LiDAR to estimate surface tilt autonomously. Additionally, the researchers suggest that more sophisticated switching rules could further optimize the transition performance. Extending this framework to dynamic environments—where the drone interacts with movable or rotating structures—remains a promising avenue for future study.
Conclusion
This study provides a rigorous foundation for stable aerial manipulation during surface interaction. By combining real-time parameter estimation with ISS-based gain scheduling, the proposed controller ensures that drones can perform complex physical tasks on varied surfaces without compromising flight stability.
Summary updated