Impedance Torque Control in Robotics

• Impedance torque control is a framework that modulates the relationship between applied torque and movement using a virtual mass-spring-damper model for compliant interaction.
• It employs nested architectures with inner torque loops and outer impedance loops enhanced by positive velocity feedback to optimize bandwidth and disturbance rejection.
• Design trade-offs in filtering, sampling, and gain tuning are critical to balancing stability, passivity, and the maximum renderable impedance (Z-width) as validated by simulation and experiments.

Impedance torque control is a methodological framework that shapes the dynamical response of a robotic joint or manipulator by modulating the relationship between applied torques and the resulting motion, typically to enforce a desired impedance—often modeled as a virtual mass-spring-damper—at a controlled interface. This paradigm is foundational for stable and safe interaction with unknown or passive environments, rendering specified mechanical behavior (compliance, stiffness, damping) at the robot–environment interface, and is achieved by nested control architectures incorporating inner torque loops, outer impedance loops, and often, feedback augmentation to enhance bandwidth or robustness. The implementation of impedance torque control has key ramifications for passivity, stability, achievable impedance range (“Z-width”), disturbance rejection, and interaction safety.

1. Core Principles and Control Architecture

Impedance torque control is structured around the enforcement of a prescribed dynamic relationship between the torque delivered at a joint and the deviation from a reference position, velocity, or force. The canonical nested architecture comprises:

• Inner Torque Loop: A PI or PID controller tracks a desired torque reference. For example, in discrete time,

$\text{PI}_t(z) = P_t + I_t \cdot \frac{z T_s}{z-1}$

where $P_t$, $I_t$ are parametrized gains (often scaled by a parameter $\beta$), and $T_s$ is the sampling time (Focchi et al., 2014).

• Outer Impedance Loop: A higher-level impedance (or position) controller, typically PD-type,

$$T_{l,\mathrm{ref}} = P_{\text{gain}} (\theta_{\text{ref}} - \theta) - D_{\text{gain}} \dot{\theta}$$

synthesizes a torque reference to render virtual stiffness ($P_{\text{gain}}$) and damping ($D_{\text{gain}}$) at the port (Focchi et al., 2014).

• Torque Plant Modeling: The actuator and transmission (including inertia, compliance, and damping) are modeled as coupled differential equations; for harmonic drives, a typical relation is

$$T_l(s) = (K_{\text{hd}} + s D_{\text{hd}}) \left[\frac{\theta_m(s)}{N} - \theta_{L_1}(s)\right]$$

with $K_{\text{hd}}$, $D_{\text{hd}}$ being gearbox stiffness and damping, and $N$ the gear ratio.

This hierarchical composition enables separation between high-bandwidth torque control and the lower-bandwidth impedance shaping, facilitating tunable compliance, disturbance rejection, and robustness against model uncertainties.

2. Enhancing Bandwidth: Positive Velocity Feedback and Transmission Zero Compensation

A defining limitation in achieving high torque loop bandwidth arises from transmission zeros induced by load and compliance dynamics, frequently situated near the system’s stability boundary (Focchi et al., 2014). The presence of these zeros impedes bandwidth increase via simple PI gain scaling due to adverse pole-zero interaction.

To mitigate this, a positive velocity feedback (“velocity compensation”) loop is introduced into the torque control. The velocity compensation gain, ideally defined as

$$\text{VC}_{\text{gain}}^{\text{ideal}}(s) = \frac{q_2(s)}{s}$$

with $q_2(s)$ a polynomial dependent on dynamics parameters, aims to cancel the deleterious transmission zeros. In practice, for tractability and noise immunity, the compensator is simplified (neglecting actuator electrical dynamics and acceleration terms) to

$$\text{VC}_{\text{gain}} = \frac{\alpha N}{k_t}(R B_m + k_t k_w),\quad \alpha > 0$$

where $\alpha$ tunes the strength of compensation. This approach shifts limiting poles towards the origin, thereby broadening the closed-loop torque bandwidth—with simulation and experimental data illustrating notably reduced torque step response times for sufficiently high $\alpha$ (Focchi et al., 2014).

However, only partial cancellation is generally possible in the presence of complex transmission zeros, and aggressive compensation must be balanced against stability and noise sensitivity constraints.

3. Filtering, Sampling, and Discrete-Time Implementation Tradeoffs

The implementation details—in particular, velocity signal filtering and digital sampling intervals—strongly affect the robust stability of the impedance-controlled system (Focchi et al., 2014). Key factors include:

• Velocity Filtering: Averaging over $N_{\text{av}}$ samples reduces quantization noise but introduces delay, enlarging instability regions especially at low stiffness values, while modestly mitigating instability at very high virtual impedance.
• Sampling Time ($T_s$): Lower $T_s$ (higher sampling rate) increases the fidelity of the digital implementation to the continuous-time dynamics. Increasing $T_s$ restricts the stable renderable impedance region (the “Z-width”), making the controller more sensitive to parameter variations and exacerbating instability risk.

Empirical validation aligns with simulation-based stability region predictions, confirming the necessity to tune filter window size, sampling rate, and controller gains as an integrated design, rather than maximizing inner-loop bandwidth in isolation.

4. Stability, Passivity, and Z-Width Characterization

Closed-loop stability and passivity are paramount for safe, robust interaction—especially when rendering high virtual stiffness/damping to passive environments. Simulation and experimental evidence reveal that:

• The interplay of inner torque loop bandwidth, impedance loop gain settings ($P_{\text{gain}}, D_{\text{gain}}$), and implementation delays determines the maximal stably renderable impedance.
• Performance metrics such as phase margin and stability region (often visualized as a “white” region signifying complete stability and “light grey” regions signifying reduced phase margin (Focchi et al., 2014)) guide gain selection.
• Although increasing torque loop bandwidth generally improves disturbance rejection, excessive bandwidth (or overly fast inner loops) can paradoxically destabilize interaction due to phase lags and unmodeled high-frequency effects.

Thus, the achievable impedance space (“Z-width”) is inextricably linked to both the control design and non-ideal digital filtering/sample time.

5. Practical Design and Deployment Considerations

Effective impedance torque control design proceeds as follows (Focchi et al., 2014):

1. Model Complete Dynamics: Accurately capture all relevant inertia, compliance, damping, and actuator electrical dynamics to identify non-minimum phase transmission zeros.
2. Inner and Outer Loop Synthesis: Select PI torque controller parameters ($\beta$), outer impedance gains ($P_{\text{gain}}, D_{\text{gain}}$), and positive velocity compensation parameter ($\alpha$) to target adequate tracking performance and bandwidth given physical limitations.
3. Filter and Sample Rate Tuning: Choose link velocity filtering window $N_{\text{av}}$ and sampling time $T_s$ to balance noise suppression and closed-loop stability, regularly corroborated through simulation and experimental region mapping.
4. Trade-off Analysis: Recognize that maximizing inner torque bandwidth is not universally optimal; the global design must maximize torque performance while maintaining the largest stable impedance rendering region, given digital limitations.
5. Validation: Use both simulation and experimental data to map and confirm the boundaries of stable and passive operation over the controller gain space.

6. Impact and Applications

The methodology allows controlled systems to achieve faster torque tracking and improved disturbance rejection with extended stability margins, provided that the design adheres to the passivity and stability constraints revealed by an integrated analysis (Focchi et al., 2014). This enables deployment in:

• Robotic manipulators interacting with unpredictable or highly variable environments.
• Human-robot interaction scenarios where safe and stable force control is paramount.
• Systems where disturbance rejection and torque tracking bandwidth must be maximized without hardware upgrades or undue hardware complexity.

The approach generalizes to a wide variety of actuator and transmission architectures by parametrizing the controller via measured/mechanically computed physical characteristics and iteratively tuning for maximized stable Z-width under application-specific timing and noise constraints.

---

In summary, impedance torque control, as characterized by a cascaded impedance/torque nested loop with velocity compensation, provides a rigorous foundation for compliant, high-performance, and robust physical human–robot/environment interaction. The design space is strongly constrained by the coupled effects of transmission zeros, sampling, and filtering, making integrated, simulation- and experiment-validated gain selection a necessity for deployment (Focchi et al., 2014).