Hybrid Force/Motion Control

• Hybrid force/motion control is a strategy that decouples motion and force regulation via selection matrices to manage robotic tasks in contact-rich environments.
• It integrates techniques from PID/PI control, switching mechanisms, and learning-based methods to ensure stability, accuracy, and adaptive performance.
• Applications span robotic assembly, surface processing, telemanipulation, and aerial tasks, enhancing safety and efficiency in uncertain or under-modeled scenarios.

Hybrid force/motion control (HFC)—also called hybrid position/force control, hybrid force/position control, or hybrid servoing—is a class of control strategies central to robotic tasks requiring simultaneous regulation of end-effector position (or velocity/trajectory) in some subspaces and contact force (or wrench) in complementary subspaces. Typical application domains include assembly, surface processing, telemanipulation, robotic learning from demonstration, and compliant interaction in both fixed-base and aerial platforms. HFC is foundational in addressing the challenges of contact-rich, under-modeled, or uncertain environments, providing robustness that cannot be achieved with pure motion or pure force control schemes.

1. Principles and Mathematical Foundations

The canonical HFC architecture decomposes the control problem into motion-controlled and force-controlled orthogonal subspaces of task space—generally via projection or selection matrices. Let $x$ denote the task-space pose, $f$ the external (measured or commanded) wrench, and $S$ a diagonal selection matrix $\in \{0,1\}^{6 \times 6}$ specifying which axes are under position control (1) and which under force control (0). The archetypal formulation produces control commands as

$$\dot{x} = S \cdot \dot{x}_{d} + (I - S) \cdot v_f, \qquad f_{cmd} = (I - S) \cdot (f_{d} - f_{measured})$$

where $\dot{x}_{d}$ is the desired velocity or position trajectory for the motion axes, and $v_f$ is determined by a force regulation law (typically PI or impedance/admittance law). This decoupling is justified under the assumption of non-conflicting kinematic and force objectives (Xie et al., 2020), and is extended to nonlinear robot dynamics by separating actuation along controllable directions (Rabindran, 2014).

The subspace split is often data-driven or optimized. For example, one can learn a time-varying constraint frame from demonstration with axes aligned to measured or desired forces, then assign force/position control dynamically along these axes (Conkey et al., 2018). In multi-contact and quasi-static manipulation, the optimal subspace split and assignment can be derived by maximizing a task-conditioned kinematic robustness criterion via closed-form or numerical optimization (Hou et al., 2020, Hou et al., 2019).

2. Architectural Variants and Modality Integration

Several HFC paradigms exist:

1. Parallel hybrid (PID/PI) design: Classic decoupled PID motion control on position axes and PI force control on force axes, possibly with off-diagonal couplings neglected (Xie et al., 2020, Yu et al., 10 Aug 2025).
2. Switching (state-based) hybrids: The controller switches between pure motion and hybrid mode based on real-time events such as contact detection (force threshold, proximity, or sensor-free proxy), with smooth transition logic to avoid instability (Liu et al., 2024, Ruderman, 2024, Kovalenko, 26 Nov 2025, Pasolli et al., 2020).
3. Unified (integral-transpose/adaptive) architectures: Single feedback law parameterized by error signals and internal state, using integral transpose-based inverse kinematics to generate seamless behavior across contact and non-contact phases, with online adaptation for model and contact uncertainties (Cos et al., 2023).
4. Optimal and learning-based hybrids: The control split and gain tuning are learned or optimized, e.g., via reinforcement learning (RL) to maximize task reward, with hybrid force–position gains as part of the action space (Wang et al., 2021).

In modern systems, HFC is embedded within hierarchical planners or learned policies, e.g., in force-centric imitation learning (Liu et al., 2024), multi-stage motion primitives for robotic assembly (Kovalenko, 26 Nov 2025), learning from demonstration with dynamic constraint frames (Conkey et al., 2018), and RL-based adaptive control with schedule-dependent subspace selection (Wang et al., 2021).

3. Implementation Details and Switching Mechanisms

Subspace Selection and Projection

The projection operators $S$, $I-S$ (or projection matrices $\Omega_f$, $\Omega_m$) are typically built from estimated normal/tangential directions (e.g., surface normal for force, its orthogonal complement for motion) (Nasiri et al., 2024). In advanced applications, the normal can be estimated online from wrench and velocity data with friction compensation, yielding adaptive subspace re-orientation in response to unmodeled surface geometry and variable friction (Nasiri et al., 2024, Guo et al., 2024).

Mode Switching and Hysteresis

Mode switching between pure position and hybrid/force mode employs either a finite-state logic (triggered by force thresholds, surface proximity, or trajectory segment flags) (Liu et al., 2024, Kovalenko, 26 Nov 2025), or continuous blending for smoothness. Hysteresis thresholds avoid chatter; dwell-time or filtered state transitions ensure stability (Praveen et al., 2020, Ruderman, 2024).

Switching may apply on:

• Events (force threshold crossing, segment flag)
• Continuous blending (smooth/softening of selection entries, e.g., exponential fade-in of force axes (Conkey et al., 2018))
• Autonomous estimator (sensor-free based on control effort overshoot (Ruderman, 2024))

Safety, Redundancy, and Null-Space Tasks

Operational space formulations permit null-space optimization (secondary objective, e.g., joint limit avoidance) while projecting primary hybrid commands into the actuated subspace (Praveen et al., 2020, Kovalenko, 26 Nov 2025).

For safety, control-barrier function (CBF)-based QP architectures enforce both force and safety constraints, guaranteeing that alignment and approach are coordinated to avoid damage (e.g., via distance–alignment constraint barriers (Chaikalis et al., 2023)).

4. Stability, Convergence, and Robustness

Hybrid force/motion control typically ensures robust regulation provided the subspace assignments do not introduce conflicts:

• Separate-loop stability: The decoupled error dynamics in each orthogonal subspace (motion/force) are designed to be stable; proofs employ standard or multiple Lyapunov functions (Xie et al., 2020, Pasolli et al., 2020, Cos et al., 2023).
• Switching stability: Stability under mode transitions is assessed using multiple Lyapunov functions, confirming monotonic decrease of the candidate in each mode and non-increase at switching (Pasolli et al., 2020, Heck et al., 2015). Sufficient damping is required to avoid oscillation (“bouncing”) on stiff environments, or compliant wrists/mechanisms must be introduced to maintain ISS bounds during reference tracking (Heck et al., 2015).
• Unified adaptive/robustness: Adaptation of link/contact parameters ensures boundedness and convergence in flexible manipulators and contact-uncertain settings, validated experimentally on embedded controllers (Cos et al., 2023).
• Contact transitions: Impact-phase or force-lag reducing feedforward/compensation laws are often included for aerial or high-velocity insertion tasks, based on instantaneous momentum transfer (Praveen et al., 2020, Guo et al., 2024).

Empirical results indicate HFC enhances compliance, accuracy, and resilience in contact-rich and under-modeled tasks, e.g., reducing contact-force peaks by up to $40\times$ in nut tightening (Kovalenko, 26 Nov 2025), $54.5\%$ relative performance improvement in contact-rich manipulation learning (Liu et al., 2024), and sub-mm accuracy in virtual fixture guidance with multi-domain actuators (Rabindran, 2014).

5. Specialized Realizations and Extensions

Force/Motion Actuator (Dual-Input)

Physical actuation architectures (FMA) implement HFC intrinsically at the actuator level, embedding dual high/low-gear inputs whose mechanical scaling (e.g., 14:1) grants selective sensitivity and authority in motion and force domains. This multi-domain strategy expands bandwidth and tracking envelope while maintaining energy efficiency under HFC blending (Rabindran, 2014).

Sensor-Free and Surface-Adaptive Hybrids

Recent progress includes loop-shaped sensor-free HFC architectures, where contact is detected via control signal deviations rather than force sensors, enabling lightweight or cost-sensitive deployments (Ruderman, 2024). Adaptive surface normal estimation and real-time friction compensation enhance tracing and surface processing on curved, uncertain workpieces (Nasiri et al., 2024).

Aerial Manipulation

In fully actuated aerial systems, HFC enables simultaneous tracking of surface-normal force and tangential motion, with explicit dynamic decoupling and stability under time-varying contact. Trajectory planning incorporates hybrid constraints and friction models, with performance validated in dynamic tasks such as aerial calligraphy (Guo et al., 2024) and inspection-on-the-fly in unknown structures (Praveen et al., 2020).

Learning-Based Hybrid Control

Imitation and reinforcement learning frameworks embed HFC as a core primitive, optimizing selection matrices, gains, and control policies. Time-varying constraint frames (Conkey et al., 2018), adaptive force-motion blending (Wang et al., 2021), and graph networks for non-rigid object shape control (Yu et al., 10 Aug 2025) have demonstrated strong data-driven generalization and robustness across diverse manipulation settings.

Port-Hamiltonian and Passivity-Based Formulations

Port-Hamiltonian unification of HFC encapsulates energy flows and power ports, enabling rigorous passivity guarantees in contact-rich and human-interactive environments. Tank-based architectures ensure that transitions between force and impedance tasks never violate passivity, preventing unsafe energetic “jumps” in settings such as ultrasound scanning or collaborative table wiping (Shao et al., 20 Oct 2025).

6. Experimental Validation and Performance Metrics

HFC architectures have been benchmarked in a variety of domains, demonstrating:

• Reduction in impact forces and overshoot at contact transitions (down to ≪1 ms detection delay, <0.05 mm overshoot (Ruderman, 2024)).
• Tight force and position tracking in contact-rich and deformable-object tasks (force error <0.15 N, position error ~4–5 mm (Cos et al., 2023, Yu et al., 10 Aug 2025)).
• Fast adaptation in learning-based systems (successful assembly/peeling >90%, completion times reduced by factors of 2–4 (Liu et al., 2024, Wang et al., 2021)).
• Energy and safety guarantees in port-Hamiltonian architectures (passivity preserved during role transitions and energetic events (Shao et al., 20 Oct 2025)).

A typical summary of implementation and results is presented in Table 1.

System/Paper	Hybrid Mechanism	Notable Performance/Outcome
(Liu et al., 2024) ForceMimic	Threshold-activated primitive	+54.5% relative success, correct force
(Kovalenko, 26 Nov 2025) Nut Tightening	Stage-based controller switching	14.5% faster, 40x force reduction
(Xie et al., 2020, Conkey et al., 2018)	Subspace-projected hybrid law	Asymptotic/BIBO stability, tight error
(Nasiri et al., 2024) Surface Normal Adaptive	On-the-fly subspace, friction comp.	<2° normal error, 5% position gain
(Guo et al., 2024) Aerial Calligrapher	Mode-mixed, force+motion decoupling	0.72N force RMSE, IoU 0.59@40cm/s
(Shao et al., 20 Oct 2025) IFIC	Port-Hamiltonian, tank energy	Passivity guaranteed, safe transitions
(Cos et al., 2023) Adaptive-Integral Unified	Smooth, adaptive law, no switches	<5mm pos. error, <0.2N force error

7. Outlook and Future Directions

Modern hybrid force/motion control is a foundational technology for robust, adaptive, and safe robot manipulation across industrial, medical, aerial, and collaborative human-interactive domains. Continued research targets:

• Data-driven and learning-based hybrid law synthesis for broader generalization and real-world adaptation.
• Port-Hamiltonian and energy-based designs for universal safety and tight passivity under uncertainty.
• Plug-and-play estimation and adaptation (surface, friction, compliance) for unstructured settings.
• Multimodal and multi-domain actuation to realize task-optimal blending of force and motion authority at both the software and hardware levels.

Advancements in closed-form optimization, adaptive projection, hybrid learning frameworks, and robust, sensor-free switching mechanisms continue to broaden the applicability and reliability of hybrid force/motion control (Hou et al., 2020, Rabindran, 2014, Kovalenko, 26 Nov 2025, Guo et al., 2024, Shao et al., 20 Oct 2025).