Passivity-based Semi-autonomous Rotational Motion Navigation for Rigid-body Networks: Stability and Human Passivity Analysis

Technical Summary: Passivity-based Semi-autonomous Rotational Motion Navigation for Rigid-body Networks

Problem Statement
While cooperative control of multi-robot systems has matured for static environments, ensuring stability in dynamic, unstructured scenarios often requires human intervention. A significant challenge remains in designing "human-in-the-loop" systems for full three-dimensional rigid-body motion (specifically attitude control on the special orthogonal group $SO(3)$). Existing approaches often struggle with conflicts between manual and autonomous control, potentially compromising closed-loop stability. Furthermore, while passivity-based methods are widely used to guarantee stability, validating the assumption that human operators behave as passive systems is non-trivial, particularly for 3D rotational kinematics. Most existing frameworks are limited to translational motion or 2D attitude kinematics.

Methodology
The paper proposes a novel passivity-based semi-autonomous attitude control framework for a network of $n$ rigid bodies interacting with a human operator. The methodology is structured as follows:

1. System Modeling: The system models $n$ rigid bodies with orientations $R_i \in SO(3)$. The collective behavior is characterized by an average vector $\bar{d}$ (derived from body-fixed vectors) and a "quasi-average" rotation matrix $\bar{R}$. The control input for each body is decomposed into an autonomous component ($\omega_{ai}$) and a human-mediated component ($\omega_{hi}$).
2. Stealthy Control Architecture: To prevent autonomous control from interfering with the human operator's perception of the collective state, the authors employ "stealthy control." The autonomous control signal is projected into the null space of the average state dynamics. This ensures that the average vector $\bar{d}$ evolves solely based on the human command, rendering the autonomous maneuvers "invisible" to the operator regarding the average state.
3. Virtual Leader and Synchronization: A virtual leader with attitude $R_l$ is introduced to mediate human intervention. The human operator commands a spatial velocity $\omega^s_h$ to synchronize a virtual leader's orientation with a reference $d_r$. The actual robot network is coupled to this virtual leader via a passivity-based attitude synchronization law.
4. Stability Analysis: The authors rigorously prove closed-loop stability under the assumption that the human operator is a passive system. They define an energy function combining the potential energy of the relative rotations and the energy of the human operator. Using LaSalle's invariance principle and Nagumo's theorem, they demonstrate that the system trajectories converge to the desired reference vector $d_r$, provided the initial relative rotation matrices satisfy specific positive semi-definiteness conditions.
5. Human Modeling and Validation: To support the theoretical analysis, the authors conducted human-in-the-loop simulations using a Virtual Reality (VR) interface (Meta Quest 3S). They identified a continuous-time transfer function model of the human operator based on experimental data where subjects aligned a virtual vector to a reference. Finally, they analyzed the passivity of these identified models using frequency-domain passivity indices.

Key Contributions
The paper outlines three primary contributions:

1. Novel Framework: Proposal of a passivity-based semi-autonomous attitude control framework specifically for human-in-the-loop rigid-body systems on $SO(3)$.
2. Rigorous Stability Proof: Establishment of closed-loop stability for the proposed architecture under the assumption of human passivity, extending previous work limited to translational or 2D kinematics.
3. Empirical Investigation: The first investigation into human passivity in the context of attitude control through human-in-the-loop simulations, involving the identification of human dynamics and the examination of passivity properties.

Results

• Theoretical: The stability proof confirms that if the human operator is passive and the initial conditions are met, the average vector $\bar{d}$ converges to the reference $d_r$ ($\lim_{t\to\infty} \|\bar{d} - d_r\| = 0$).
• Simulation and Identification: A human operator model was successfully identified from simulation data (using 12 trials of 180 seconds total). The model, a 2-by-2 transfer function matrix, captured the dominant input-output behavior with fit ratios ranging from ~51% to ~77% depending on the coupling gain $k_s$.
• Passivity Analysis: The analysis of the identified models revealed that human passivity is sensitive to system parameters. Specifically, for a coupling gain $k_s = 3.0$, the identified model satisfied the passivity condition ($\nu(\omega) \ge 0$) across the frequency spectrum. However, for $k_s = 2.5$, the passivity condition was violated at frequencies above approximately 2 rad/s.

Significance and Claims
The paper claims to address a critical gap in the literature by establishing a stability framework for full 3D rigid-body motion under human intervention. The authors emphasize that while their theoretical framework relies on the assumption of human passivity, their experimental results highlight that this assumption is not universally guaranteed; it depends heavily on parameter tuning (specifically the coupling gain $k_s$). Consequently, the paper modestly concludes that while the proposed framework is stable under ideal passive assumptions, future work must extend the design to accommodate "passivity-short" human operators to ensure robustness in practical applications. The work is presented as a preliminary step toward solving the broader challenge of 3D rigid-body motion control in semi-autonomous networks.