Actuator Side Control

• Actuator side control is defined as relocating control computations to or near actuator nodes, enhancing performance and robustness in uncertain environments.
• It employs distributed algorithms such as dynamic controller placement, adaptive allocation, and sensorless state estimation to mitigate communication and model uncertainties.
• This approach improves closed-loop stability and rapid response across applications like soft robotics, aerospace systems, and industrial wireless networks.

Actuator side control refers to the suite of architectures, algorithms, and system-level strategies in which key control computations and system adaptation are placed at or near the actuator, as opposed to being centralized with the main controller or collocated with sensor-side processing. This paradigm encompasses methodologies for increasing resilience, performance, and flexibility in the presence of uncertainties, disturbances, or variable communication conditions, especially in multi-node or networked control settings. Recent research has investigated its importance and practicality across a spectrum of domains, including wireless sensor–actuator networks, soft robotics, electromechanical systems, and large-scale distributed control.

1. Architectural Principles of Actuator Side Control

A distinguishing characteristic of actuator side control is that the control computation—ranging from basic feedback laws to state estimation and even adaptation—is co-located with or closely follows the actuator node in the plant’s signal path. While traditional architectures tie sensing, computing, and actuation into rigid node roles, actuator side control relaxes these designations. For wireless sensor–actuator networks, nodes dynamically assume controller duties depending on network state and real-time transmission outcomes, as formalized in the adaptive controller placement architecture (Quevedo et al., 2013). Each node along the data-relay path computes a local state estimate and proposes a tentative control input; the final actuation value is determined by which node’s computation survives the network’s erasure process.

In multi-actuator and over-actuated systems, actuator side control often means embedding allocation, adaptation, or redundancy management directly at the actuation interface. Control allocation frameworks distribute global control signals to individual actuators, dynamically compensating for actuator degradation or redundancy, as in discrete adaptive control allocation for over-actuated sampled-data systems (Tohidi et al., 2021).

2. Adaptation to Network and System Uncertainties

Robustness to communication uncertainties is a central motivation for actuator side control in wireless environments. In erasure-channel networks, dynamic allocation of the controller role allows the system to adapt in real time to stochastic dropouts. The controller’s geographic location is a function of forward and feedback link realizations (captured by Bernoulli or Markovian models), allowing nodes with more precise or up-to-date state estimates to assume the control computation in the event of data loss at downstream nodes (Quevedo et al., 2013).

Actuator side control is also used to manage physical model uncertainties, such as those arising from unmodeled dynamics or parameter variability. For instance, in sensorless control of antagonistic PAM actuators (Shin et al., 2021), an unscented Kalman filter runs local state estimation at the actuator, using only pressure sensors to estimate both torque and stiffness, while in deep learning-enhanced electromechanical actuators, physics-informed surrogates correct discrepancies between analytical models and observed actuator outputs, leading to reliable sensorless control (Bahari et al., 19 Sep 2025).

3. Control Laws, Estimation, and Computation

Implementation of actuator side controllers demands lightweight, distributed algorithms suited for limited computational resources and variable feedback. Distributed architectures leverage local computation and selective relay of tentative control inputs, as in Algorithm 1 of (Quevedo et al., 2013). State estimation and control synthesis must be robust to missing or delayed data; e.g., nodes update both local plant output estimates and tentative control values using history and current packet status.

In the presence of functionally redundant actuators or complex dynamics, actuator-side algorithms perform local adaptation or allocation. Discrete adaptive control allocation—utilizing closed-loop reference models and online adaptation of a parameter matrix—enables allocation of the total control effort spectrally among actuators with bounded uncertainties, while guaranteeing closed-loop stability and rapid convergence, without explicit identification of actuator effectiveness (Tohidi et al., 2021). This paradigm, further, does not require persistency of excitation or uncertainty estimation, but relies on a normalized error signal and parameter update law.

Sensorless actuator side control further integrates state estimation at the actuation interface: surrogate models trained on physics plus data (e.g., Kriging in (Bahari et al., 19 Sep 2025)) transform observable motor-side variables into accurate predictions of load force and velocity, enabling voltage-command synthesis in hierarchical VDC frameworks without dedicated load-side sensors.

4. Performance Metrics and System Evaluation

Actuator side control is assessed primarily by closed-loop cost indices reflecting control performance, stability, and responsiveness under uncertainties. For the adaptive controller placement in erasure channels, closed-loop quadratic cost (e.g., $J = \sum_{k} y_k^2$) is minimized compared to fixed-node architectures, confirmed both in analysis (via stationary state covariance of jump-linear systems) and in simulations (Quevedo et al., 2013). The empirical distribution of controller locations under i.i.d. dropout processes highlights how control allocation shifts closer to sensors as network conditions degrade, mitigating the adverse effect of poor state estimates at the actuator under high loss rates.

In over-actuated systems, convergence rates and closed-loop stability are critical. The adaptive control allocator guarantees boundedness and asymptotic convergence of allocation error without oscillatory transients (Tohidi et al., 2021). Numerical experiments with the ADMIRE aircraft model demonstrate that, when facing actuator effectiveness drops, the approach sustains tight tracking and rapid adaptation.

In sensorless heavy-duty manipulators, position, velocity, and force tracking errors are analyzed under varying payload conditions; sensorless control achieves comparable accuracy to feedback-based control, validating the efficacy of local estimation and actuation (Bahari et al., 19 Sep 2025).

5. Practical Considerations, Overhead, and Limitations

The communication and computational overhead of actuator side control is a crucial design factor. For the adaptive controller-placement method, packets are augmented with a single tentative control field, ensuring minimal overhead—two fields per packet—without requiring network acknowledgments or explicit time-based role scheduling (Quevedo et al., 2013). This design enables deployment in energy- or resource-constrained environments, typical of wireless sensor–actuator networks or industrial IoT settings.

Potential limitations stem from assumptions of local computation feasibility, sufficient feedback mechanisms, and the statistical independence of transmission outcomes. In architectures relying on dynamic controller relocation, practical implementation requires synchronization to determine the surviving tentative control field and adequate feedback links to propagate prior control signals upstream.

Functionality under model uncertainty demands both identification schemes and robust adaptation. The quality of estimation and adaptation is tied to the accuracy of the local process model (as in unscented Kalman filtering for PAMs (Shin et al., 2021)) or the representational capability of data-driven surrogates in sensorless frameworks.

6. Applications and Case Studies

Actuator side control methodologies have been validated across diverse real-world systems:

• Wireless process control: Flexible networked control in industrial wireless sensor–actuator networks, where dynamic controller assignment yields improved recovery and lower cost indices under varying packet dropout rates (Quevedo et al., 2013).
• Over-actuated aerospace systems: Adaptive control allocation on the ADMIRE platform demonstrates fault-tolerant allocation under actuator degradation with rapid convergence (Tohidi et al., 2021).
• Heavy-duty robotic manipulators: Unified frameworks integrating actuator configuration optimization with sensorless control enable precise and efficient actuation even with large payloads, confirmed in hardware-in-the-loop experimentation (Bahari et al., 19 Sep 2025).
• PAM soft robotics: Sensorless angle and stiffness control using only pressure sensors provides lightweight actuation for compliant limbs and exoskeletons (Shin et al., 2021).
• Collaborative multi-actuator HCI: Dynamic allocation strategies increase system responsiveness while maintaining safety constraints by minimizing actuation delay and ensuring bounded outputs.
• Wireless over-the-air control: Computation of control signals directly at the actuator via concurrent sensor transmissions (exploiting wireless channel superposition) drastically reduces actuation delay and expands the stability region in networked closed-loop control (Park et al., 2021).

7. Implications for System Design

Actuator side control challenges the traditional fixed role assignment of controllers and highlights the importance of adaptive, distributed architectures in modern networked and robotic systems. By enabling dynamic relocation and local adaptation of control calculations, actuator side control mitigates the inherent performance–robustness trade-offs due to unreliable communication, incomplete information, and system uncertainties.

The practical takeaways are clear: distributed architectures with minimal communication overhead and real-time local estimation enable robust, scalable control with improved cost, stability, and recovery characteristics in both wireless and hard-wired environments. These findings inform future design of resilient networked control systems and constitute the foundation for next-generation autonomous, fault-tolerant, and resource-efficient actuation.