Task-Space Admittance with S–M–D Dynamics

• The paper demonstrates that task-space admittance via S–M–D dynamics enables robots to mimic physical mass–spring–damper responses for robust and compliant force control.
• It presents a rigorous mathematical formulation and adaptive control strategies that tune virtual mass, damping, and stiffness parameters to handle variable payloads and environmental uncertainties.
• Practical insights and experimental validations illustrate how constraint-based admittance and mechanical synthesis enhance safety in human–robot interactions and surgical teleoperation.

Task-space admittance via spring–mass–damper (S–M–D) dynamics is a foundational methodology in robot force control whereby a manipulator’s end-effector is commanded to emulate the dynamic response of a physical mass–spring–damper system in Cartesian space. This compliance shaping is critical to achieving robust, predictable, and safe interaction with manipulated payloads and uncertain or dynamic environments. By explicitly synthesizing an admittance mapping, robotic systems admit (respond to) interaction wrenches by generating prescribed trajectories in task space, with the effective inertia, damping, and stiffness set by virtual parameters and possibly updated in real time. The architecture is central to applications such as adaptive manipulation, human-robot physical interaction, surgical teleoperation with spatial constraints, and the synthesis of physical or virtual compliant behaviors.

1. Mathematical Formulation of Task-Space S–M–D Admittance

The canonical task-space S–M–D admittance law prescribes the desired mapping between the external wrench $\bm F_{\rm ext}\in\mathbb{R}^3$ and the end-effector’s motion in Cartesian position $\bm p_a\in\mathbb{R}^3$: $$\bm F_{\rm ext}(t) = M_a\left(\ddot{\bm p}_a - \ddot{\bm p}_0\right) + B_a\left(\dot{\bm p}_a - \dot{\bm p}_0\right) + K_a\left(\bm p_a - \bm p_0\right) \tag{1}$$ where $M_a, B_a, K_a\in\mathbb{R}^{3\times3}$ are diagonal “virtual” mass, damping, and stiffness matrices, and $\bm p_0$ is the reference pose. This law admits exact inversion to yield the virtual acceleration: $$\ddot{\bm p}_a = M_a^{-1} \left( \bm F_{\rm ext} - B_a\dot{\bm p}_a - K_a(\bm p_a-\bm p_0) \right) \tag{2}$$ Numerical integration produces the commanded velocity signal: $$\dot{\bm p}_a(t) = \int_{t_0}^t \ddot{\bm p}_a(\tau)\,d\tau \tag{3}$$ which is relayed to the velocity-tracking loop of the manipulator (Gholampour et al., 22 Apr 2025, Li et al., 2016, Scherzinger et al., 2020, Haninger et al., 2022, Kastritsi et al., 2022).

This scheme ensures that the end-effector admits any external force as if it were attached to a notional mass–spring–damper, leading to a transfer function: $X(s) = (M_a s^2 + B_a s + K_a)^{-1} F_{\rm ext}(s)$ whose dynamic shape is controlled entirely by the choice of $M_a,\, B_a,\, K_a$.

2. Adaptive and Enhanced Control Architectures

Fundamental S–M–D admittance can be extended by embedding adaptive estimation, environmental modeling, or constraints:

• Adaptive Mass Estimation for Unknown Payloads: For manipulation with varying object mass, the closed-loop system augments the admittance with an excitation force $\bm F_{\rm exc}$, estimated online using

$$m_u = \frac{f_z}{\ddot p_z} - m_g \qquad \bm F_{\rm exc} = m_u (\ddot p_z)\, \hat{\bm z} \tag{8, 9}$$

where $m_g$ is the gripper mass and $f_z$ the FT sensor reading along gravity, feeding back the total estimated mass into the control to eliminate vertical “sag,”

$\delta_z = \frac{m g}{K_{zz}} \tag{4}$

and preserve compliance while compensating for variable payloads (Gholampour et al., 22 Apr 2025).

• Simultaneous Adaptation of Environmental Impedance: When environmental stiffness and damping are unknown or time-varying, an adaptive control law updates estimates using

$$\delta K_S(t) = Q_S [ \varepsilon(t)x(t)^{T} - \beta(t)K_S(t) ], \quad \delta K_D(t) = Q_D [ \varepsilon(t)\dot x(t)^{T} - \beta(t)K_D(t) ]$$

with Lyapunov-backed stability proofs (Li et al., 2016).

• Constraint-Based Admittance with Safety Enforcement: Applications such as surgical teleoperation require admittance to be imposed in a subspace with additional safety constraints. The admittance equation is split:

$$M_d\ddot \xi + D_d\dot \xi + K_d\xi = F_{\rm th} + F_r$$

where compliance is enforced in “free space,” while forbidden regions are encoded by artificial potential fields that produce repulsive wrenches, guaranteeing both bounded tool motion and strict passivity under human actuation (Kastritsi et al., 2022).

• Virtual Forward Dynamics for Enhanced Manipulability: By synthesizing a virtual rigid-body model with dominant end-effector inertia ($\gamma = m_e / m_l \gg 1$), the operational-space inertia $\Lambda(q) \approx m_e I_6$ yields a decoupled and tunable S–M–D law, integrating advantages of both Jacobian-inverse and -transpose approaches but with improved stability near kinematic singularities (Scherzinger et al., 2020).

3. Parameter Synthesis and Tuning Guidelines

Empirical and analytical selection of $(M_a, B_a, K_a)$ is central for achieving desired compliance, stability, and responsiveness:

• Virtual Mass $M_a$: Should approximate or slightly exceed the expected total load (end-effector + payload). Typical experimental values ranged from $m=4$ kg for a 1.5 kg payload (Gholampour et al., 22 Apr 2025) to $M_d=13$–$20$ kg for high-payload collaborative tasks (Haninger et al., 2022). In virtual forward dynamics, $m_e$ is maximized (linked-mass ratio $\gamma\gtrsim 10^3$) to decouple motions across axes (Scherzinger et al., 2020).
• Stiffness $K_a$: Directly controls static compliance. For vertical gravity compensation,

$k = \frac{mg}{\delta_z}$

is set according to the allowable sag. High stiffness reduces tracking errors but increases transmitted forces in collision. Typical $k$ in the range $[100,2500]$ N/m; approaches with $K_d = 0$ are preferred if compliance is paramount (Gholampour et al., 22 Apr 2025, Haninger et al., 2022).

• Damping $B_a$: Crucial for stability and overshoot minimization. Critical damping,

$b = 2\sqrt{m k} \tag{10}$

is used routinely. For high-payload systems, $B_a$ must be set to exceed the combined environmental and system lower bounds, e.g., $B_a = 1000$–$2000$ N·s/m for 16–50 kg loads (Haninger et al., 2022).

Feedback and feedforward augmentation—lead compensation, direct feedback damping, and (optionally) payload-acceleration cancellation—further shape dynamic response and improve robustness with respect to environmental contacts and compliance variability (Haninger et al., 2022).

4. Passivity and Stability Guarantees

S–M–D admittance controllers are designed to be strictly passive mappings from external wrench to velocity (or position). The system

$$M_a \ddot x + B_a \dot x + K_a(x-x_0) = F_{\rm ext}$$

is passive and stable for $M_a, B_a, K_a \succeq 0$. For systems with environmental contact, strict passivity critically depends on sufficient admittance damping, and compliance is often added mechanically to the environment (e.g., leaf springs or compliant tables) to further reduce peak forces (Haninger et al., 2022).

When adaptive estimation (mass, stiffness, damping) is used, Lyapunov-based analyses are employed to guarantee ultimate boundedness of tracking errors and parameter estimates. In all cited cases, energy storage functions or composite Lyapunov functionals are constructed such that their derivative is negative semi-definite, ensuring BIBO stability (Li et al., 2016, Gholampour et al., 22 Apr 2025, Kastritsi et al., 2022).

In constraint-admittance architectures, passivity in the unconstrained (“free”) subspace is strictly preserved, while constrained subspaces are stabilized by S–M–D dynamics with no input energy inflow (Kastritsi et al., 2022).

5. Physical Realization, Mechanical Circuit Synthesis, and Embedding

Mechanical realization of desired admittance behaviors is possible by synthesizing series–parallel networks of physical springs, dampers, and inerters. For any target third-order positive-real admittance $Y(s)$,

$$Y(s) = \frac{b_3 s^3 + b_2 s^2 + b_1 s + b_0}{s(a_2 s^2 + a_1 s + a_0)}$$

Wang, Chen, and Liu (Wang et al., 2023) provide a cook-book approach to

• Test for positive-realness via Bezoutian criteria,
• Select one of twelve canonical series–parallel (up to six-element) topologies,
• Compute each element’s value (spring $k_i$, damper $c_i$, inerter $b_i$) in closed form,
• Physically instantiate the network (e.g., as a “inerter box”) interfaced to a robotic force actuator,

thus imposing the transfer function $F(s) = Y(s)V(s)$ at the end effector.

This exact realization ensures passivity, minimal element complexity, and straightforward Bode-plot verification, directly implementing a virtual S–M–D law in physical hardware (Wang et al., 2023).

6. Experimental Validation and Performance

Experimental validations across multiple studies provide quantitative evidence of efficacy in real manipulation tasks:

• Mass-Adaptive Payload Handling: For a UR5e lifting 1.5 kg, a low-stiffness admittance controller with adaptive mass compensation accomplished successful pick-and-place tasks, controlled end-effector sag below tight tolerance ($\delta_z < 3.5$ mm, RMSE = 1.99 mm), and avoided gross instability even with significant virtual compliance. Without compensation, the same stiffness led to large errors ($\delta_z = 46.8$ mm, RMSE = 20.58 mm) and task failure (Gholampour et al., 22 Apr 2025).
• High-Payload Co-Manipulation and Contact Tasks: In COMAU Racer7 (16 kg) and AURA (50 kg) manipulators, task-space admittance shaped by $M_a = 13$–$20$ kg, $B_a = 1000$–$2000$ N·s/m enabled precise, responsive, and robust manual guidance with successfully bounded contact forces and reduction of collision-induced spikes by up to 42% via a “zero-velocity-reset” scheme and environmental compliance (Haninger et al., 2022).
• Constraint Enforcement in Surgical-Type Tasks: Passive S–M–D-based admittance controllers with subspace splitting enabled precise enforcement of remote center-of-motion (RCM) and forbidden-region avoidance, demonstrated by sub-millimeter RCM tracking and provable safety in KUKA LWR4+ setups (Kastritsi et al., 2022).
• Adaptive Environment Identification: Simulations reveal that S–M–D admittance control with adaptive impedance estimation accurately tracks time-varying environmental parameters, with convergence to true values within <5% in under 2 seconds and tracking RMS error below 0.5 mm (Li et al., 2016).

7. Practical Insights, Limitations, and Design Guidance

Design and implementation of task-space S–M–D admittance require adherence to stability and performance guidelines:

• Virtual inertia and damping must be tuned to payload and environmental properties, with an emphasis on matching $M_a$ to payload inertia and $B_a$ exceeding thresholds set by environmental stiffness.
• High compliance (low stiffness) should be paired with real-time gravity/mass compensation to avoid loss of accuracy or instability when payload is unknown or variable.
• For safe human–robot interaction and task execution involving physical contact, auxiliary passive compensation (e.g., mechanical compliance) and damping feedback loops are essential for passivity and bounded force response.
• When constraint or environmental uncertainty is dominant, adaptive estimation (of mass, stiffness, and damping) and barrier formulations (for constraint enforcement) significantly enhance robustness and safety.
• Mechanical synthesis provides a straightforward path toward physically realized virtual S–M–D admittance, ensuring passivity and facilitating critical applications demanding hardware-level compliance.

Taken together, these techniques render task-space spring–mass–damper admittance the principal tool for explicit, physically-intuitive, and stable compliance shaping in advanced robotic manipulation, flexible automation, and safe human–robot interaction (Gholampour et al., 22 Apr 2025, Li et al., 2016, Scherzinger et al., 2020, Haninger et al., 2022, Wang et al., 2023, Kastritsi et al., 2022).