Understanding and Utilizing Dynamic Coupling in Free-Floating Space Manipulators for On-Orbit Servicing
Authors: Das, Choi, Kim · Year: 2025 · Venue: AAS (AAS conference paper; uses the AAS template)
Raw: md
Summary
The paper proposes a dynamic-coupling-informed trajectory optimization scheme for a free-floating space manipulator system (SMS). Rather than treating base-manipulator dynamic coupling as a disturbance to suppress, it exploits coupling to assist end-effector motion with no base actuation. A singular value decomposition (SVD) of the joint-to-base coupling matrix (C_{bm}) extracts dominant coupling directions; two scalar metrics (normalized Shannon entropy of the singular-value spectrum and a coupling-assistance cosine) drive the trajectory cost. A sliding mode controller (SMC) then tracks the optimized joint trajectory and is shown Lyapunov-stable.
Key Claims
- The joint-to-base coupling matrix (C_{bm} \in \Re^{6\times n}) (Eq. 24), obtained by inverting the locked momentum-conservation block, fully maps joint rates to base linear/angular velocities under zero initial momentum; its SVD’s first pair ((v_1,u_1)) is the dominant coupling direction and the last pair approaches the base near-null-space when (\sigma_n \approx 0).
- Two coupling quality metrics are introduced: normalized Shannon entropy (H_{norm}\in[0,1]) of the normalized singular-value spectrum (Eq. 29; 0 = highly directional/single-mode coupling, 1 = uniform) and a coupling-assistance cosine (\cos(\theta_a)\in[-1,1]) (Eq. 31; +1 fully assists desired EE motion, -1 opposes).
- Joint velocities are parameterized as a weighted sum of right-singular vectors, (\dot q_{DC}=\sum_i \alpha_i v_i) with (\alpha_i\in[-1,1]) (Eq. 30); the optimizer selects the (\alpha_i) minimizing the cost in Eq. 32 subject to joint angle/rate/acc limits, inter-link and link-base collision avoidance, and a terminal EE-error bound active over the last 10% of the horizon (Eqs. 33-40).
- Demonstrated on a planar 3-link single-arm SMS (base 31.015 kg; three 0.569 kg links): short-range task keeps net base displacement bounded and minimal; long-range task deliberately recruits larger (but bounded) base motion to reach a distant target.
- The SMC tracking torque (Eq. 43) uses the reduced manipulator inertia (H_q) (Eq. 44, the Schur complement of the base block) and reduced Coriolis term (C_q); Lyapunov analysis gives (\dot V_r \le -K_s s^T \mathrm{sat}(s/\lambda)) and exponential convergence on the sliding surface, with the caveat that (\dot H_q) is neglected.
Method
Regime: strictly free-floating. The base receives no thruster/wheel control: “no control applied to the satellite base,” and the first six entries of the generalized force (\tau) are zero. (The paper carefully contrasts this with free-flying systems, where the base is actively controlled in translation and rotation — that distinction is central to its motivation, since their algorithm exists precisely to operate when base actuation is unavailable, e.g. close-proximity ops or actuator failure.)
Model: rigid (n)-DOF arm on a 6-DOF base, quaternion base attitude kinematics (\dot\epsilon = \tfrac12 G(\epsilon)\omega_b) (Eq. 1-2). EE velocity (Eq. 3, 7) splits into base Jacobian (J_b) and manipulator Jacobian (J_m). Lagrangian = kinetic energy (microgravity, (V\approx 0)) gives the partitioned mass matrix (Eq. 9) with blocks (H_v, H_{v\omega}, H_\omega, H_{vm}, H_{\omega m}, H_m) (Eqs. 10-15), and EOM (H(\Phi)\ddot\Phi + C(\dot\Phi,\Phi)\dot\Phi = \tau) (Eq. 19).
Zero-momentum conservation (Eqs. 20-22) yields the coupling map
[
\begin{bmatrix}\dot r_b\ \omega_b\end{bmatrix} = C_{bm}\dot q,\quad
C_{bm} = -\begin{bmatrix}H_v & H_{v\omega}\ H_{v\omega}^T & H_\omega\end{bmatrix}^{-1}\begin{bmatrix}H_{vm}\ H_{\omega m}\end{bmatrix}.
]
Substitution gives the generalized Jacobian (J^* = J_b C_{bm} + J_m) (Eq. 25) and the end-to-base coupling (C_{be}=C_{bm}J^{*\dagger}) (Eq. 27).
Notation flags for the pencil-and-paper reader:
- (\Gamma \in \Re^{3\times3}) (Eq. 42) is fixed at 3×3 even though the sliding surface is over (n) joints; for the (n=3) demo this is fine but the dimension does not generalize as written.
- (\tilde C = \tfrac{1}{\kappa}\log(1+\exp(\kappa\cos\theta_a))) (a softplus of the assistance cosine) appears in cost Eq. 32 but (\kappa) is never numerically specified.
- The entropy formula (Eq. 29) is borrowed from an ecology paper (ref. 22); (s_i=\sigma_i/\sum_j\sigma_j).
Relevance to thesis
Direct counterpoint and complement to our free-flying program. We have a fully-actuated base; this paper deliberately discards base actuation and shows how much task progress can be extracted from coupling alone. Useful as (a) a degraded/failure-mode baseline (“what if the base actuators die mid-task”), (b) a source of the SVD-of-(C_{bm}) coupling-direction analysis we can repurpose for coordinated whole-body planning, and (c) a concrete softplus/entropy cost design we could adapt — or critique — for our risk-aware layer. The collision and terminal-tolerance constraint set is a clean template for our motion-planning formulation.
Connections
Topics: dynamic_coupling · generalized_jacobian · free_flying_vs_free_floating · sliding_mode_control
Key Equations / Quotes
“An SMC is designed to track the optimized trajectory using joint torques, with no control applied to the satellite base.” (p. 2)
“By leveraging the dynamic coupling effect rather than opposing it, free-floating SMSs can perform precise manipulations while preserving system integrity and resource efficiency.” (p. 2)
Coupling matrix (Eq. 24):
[
C_{bm} = -\begin{bmatrix}H_v & H_{v\omega}\ H_{v\omega}^T & H_\omega\end{bmatrix}^{-1}\begin{bmatrix}H_{vm}\ H_{\omega m}\end{bmatrix}.
]
Generalized Jacobian (Eq. 25): (\begin{bmatrix}\dot r_e\ \omega_e\end{bmatrix} = (J_b C_{bm} + J_m)\dot q = J^*\dot q.)
Normalized SVD entropy (Eq. 29): (H_{norm} = -\dfrac{\sum_{i=1}^n s_i \log s_i}{\log n},\quad s_i=\sigma_i/\textstyle\sum_j \sigma_j.)
Coupling assistance (Eq. 31): (\cos(\theta_a) = \dfrac{\hat d^T \dot r_{e_{DC}}}{|\hat d|,|\dot r_{e_{DC}}|}.)
Cost (Eq. 32): (J = \sum_{k=1}^N \big(\tilde C(k) - (1 - H_{norm}(k))\big)
Reduced manipulator inertia (Eq. 44, Schur complement):
[
H_q = H_m - \begin{bmatrix}H_{vm}\ H_{\omega m}\end{bmatrix}^T \begin{bmatrix}H_v & H_{v\omega}\ H_{v\omega}^T & H_\omega\end{bmatrix}^{-1}\begin{bmatrix}H_{vm}\ H_{\omega m}\end{bmatrix}.
]
Open Questions
- The cost Eq. 32 couples a softplus of (\cos\theta_a) against ((1-H_{norm})); what does the minimizer mean physically, and how sensitive is it to the unspecified (\kappa) and to the entropy normalization choice?
- (\Gamma) is declared 3×3 (Eq. 42); how does the surface/controller generalize to (n>3) or to the 3-D (non-planar) arm promised as future work?
- Lyapunov analysis neglects (\dot H_q) “under the assumption that it is bounded and relatively small” — is this assumption justified during the large, fast base motions seen in the long-range case?
- Validation is simulation-only on a planar single-arm system; the cited air-bearing hardware testbed results are not yet reported.
- No quantitative comparison against a coupling-minimizing or free-flying baseline is given, so the claimed efficiency/energy advantage is asserted rather than measured.