Task-Adaptive Admittance Control for Human-Quadrotor Cooperative Load Transportation with Dynamic Cable-Length Regulation

1. Problem Statement

The paper addresses the safety and efficiency challenges in Human-Quadrotor Cooperative Load Transportation (CLT), specifically where a human operator physically guides a suspended load attached to a drone via a cable.

• Current Limitations: Existing research in Physical Human-Robot Interaction (pHRI) focuses heavily on robotic manipulators. In aerial systems, most approaches use aerial manipulators (arms) or simplified cable models that ignore the coupled dynamics between the drone, the cable, and the swinging load.
• Specific Challenges:
  • Safety: Unstructured environments pose risks of load collisions or unpredictable drone behavior during human interaction.
  • Dynamic Complexity: Traditional controllers often treat the drone and load as decoupled systems or rely on fixed cable lengths, failing to account for the complex interaction forces and the benefits of actively adjusting the cable length.
  • Adaptability: Existing methods often require explicit knowledge of load mass/size or rely on external human commands, limiting autonomy and flexibility.

2. Methodology

The authors propose a novel control framework centered on a Coupled Virtual Impedance Model (CVIM) integrated with an actively controlled winch.

A. System Modeling

• Hardware: A quadrotor equipped with an actuated winch, a load cell, and a suspended load.
• Dynamics: The system is modeled using coupled differential equations that account for the quadrotor's translational/rotational dynamics, the cable's angular velocity, and the load's inertia.
• Key Insight: The model treats the contact force ($F_c$) exerted by the human on the load hook as the primary input. Crucially, the model decouples the effects of $F_c$ into rotational components (cable inclination angle $\beta$) and translational components (cable length $L$).

B. Control Strategy: Coupled Virtual Impedance Model (CVIM)

The core innovation is an admittance controller that shapes the system's response to human forces using a virtual mass-spring-damper model:

1. Virtual Impedance: The controller defines a desired compliant behavior where deviations in cable inclination ($\delta\beta$) and length ($\delta L$) from a virtual reference trajectory follow a second-order mechanical dynamic equation:
   $$M\ddot{\zeta} + B\dot{\zeta} + K\zeta = \tau_c$$
   Where $\zeta = [\delta\beta, \delta L]^T$, and $M, B, K$ are inertia, damping, and stiffness matrices derived from the system's physical parameters.
2. Command Shaping: To ensure the generated motions are physically trackable by the drone, the controller uses a "command-shaping" law. This combines a proportional term (for immediate responsiveness) with a discounted integral term (to smooth out rapid force changes and prevent aggressive maneuvers).
3. Output: The controller outputs desired adjustments to the quadrotor's position ($\delta x_q$) and the cable length ($\delta \bar{L}$), allowing the drone to move and the winch to reel in/out simultaneously to maintain the desired interaction profile.

C. Baseline Comparison

To validate the CVIM, the authors developed a Simplified Virtual Impedance Model (SVIM) as a benchmark. The SVIM treats the drone's translational motion and cable length variation as separate, decoupled systems, ignoring the coupled rotational dynamics of the cable.

3. Key Contributions

1. First Active Cable-Length System for CLT: Introduction of the first quadrotor system designed specifically for human-quadrotor CLT with an actively controlled winch, enabling dynamic cable length regulation.
2. Coupled Dynamics Modeling: Development of a versatile dynamic model that accounts for the coupled interaction between the drone, the swinging load, and the cable, without requiring prior knowledge of the load's physical attributes (mass/size).
3. Novel Admittance Controller (CVIM): A control strategy that unifies cable inclination and length regulation within a single virtual impedance framework, effective under both constant (CCL) and varying (VCL) cable length conditions.
4. Comprehensive Experimental Validation: Rigorous testing across two distinct tasks (in-place loading/unloading and obstacle avoidance during transport) comparing CVIM vs. SVIM under varying stiffness and cable configurations.

4. Experimental Results

Experiments were conducted in an indoor motion-capture environment with a human operator guiding the system through two tasks: Loading/Unloading and Transporting (with obstacle avoidance).

• Performance Metrics: Traversed distance ($d$), average cable inclination ($\bar{\beta}$), trajectory smoothness (jerk, $\bar{J}_q$), and average tensile force ($\bar{f}_T$).
• Key Findings:
  • Responsiveness: The CVIM controller significantly increased the traversed distance ($d$) compared to SVIM, indicating better compliance with the human operator's guidance.
  • Stability & Safety: CVIM combined with Variable Cable Length (VCL) resulted in significantly smaller cable inclination angles ($\bar{\beta}$), reducing load swing.
  • Smoothness: CVIM produced smoother trajectories (lower jerk, $\bar{J}_q$) compared to SVIM, particularly at high stiffness levels.
  • Force Reduction: The proposed method reduced the average interaction force ($\bar{f}_T$) required by the human operator, making the system easier to control.
  • Synergy: The combination of the CVIM controller and dynamic cable length regulation (VCL) showed synergistic effects, outperforming both the simplified controller and fixed-length configurations across all metrics.

5. Significance and Conclusion

This work represents a significant step forward in safe human-aerial robot collaboration.

• Safety: By actively regulating cable length and accounting for coupled dynamics, the system minimizes load swing and interaction forces, reducing the risk of collisions in unstructured environments.
• Efficiency: The system allows for faster and more fluid cooperative transport, as the drone naturally adapts to the human's pace and direction without requiring complex force estimation algorithms or explicit load parameterization.
• Future Impact: The framework provides a robust foundation for last-mile delivery and infrastructure inspection where human-drone interaction is necessary. The authors note future work will focus on removing reliance on external motion capture (using onboard sensors) and handling more complex collision scenarios.

In summary, the paper demonstrates that task-adaptive admittance control grounded in coupled dynamics significantly outperforms conventional decoupled approaches, enabling safer, smoother, and more efficient human-quadrotor cooperative load transportation.