Time-Varying Kinematic Control in Robotics

Updated 9 April 2026

Time-Varying Kinematic Control is a methodology that designs feedback controllers for robotic systems with time-dependent objectives, constraints, and trajectories.
Key techniques such as Constrained SLQ, LTV–LQR, and adaptive control barrier functions systematically address real-time safety, stability, and performance in dynamic environments.
These approaches enable prescribed-time convergence, formation control, and learning-based adaptations, making them effective for mobile manipulators, multi-agent systems, and teleoperation.

Time-varying kinematic control refers to the design, analysis, and implementation of feedback controllers for robotic systems whose control objectives, reference signals, and/or constraints explicitly depend on time. In contrast to time-invariant settings, time-varying kinematic control frameworks address challenges such as time-dependent task-space trajectories, time-varying constraints, formation morphing, safety in dynamically changing environments, and prescribed-time stabilization. This area plays a central role in modern robotics, encompassing manipulators, mobile robots, multi-agent systems, and teleoperation platforms where flexibility, adaptability, and temporal specification are essential.

1. General Principles and Problem Formulation

Time-varying kinematic control is fundamentally concerned with systems governed by equations of the form

$\dot x(t) = f(x(t), u(t), t),\quad x\in\R^n,\, u\in\R^m,$

where $f$ may be time-dependent, and the control objective can include tracking a time-varying reference $x_d(t)$ , enforcing time-dependent constraints, or achieving global stabilization within a specified time-window. The tracking or regulation cost is typically formulated as

$J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$

subject to explicit equality and/or inequality constraints that may themselves depend on time or both state and time, e.g., $h(x, u, t) = 0$ (Giftthaler et al., 2017).

Key challenges include:

Explicit temporal variation in dynamics, references, and/or task-space constraints.
Accommodation of nonholonomic (velocity) and holonomic (position/operational-space) constraints, which may themselves change over time.
Stability, performance, and feasibility in the presence of time-dependent domains, uncertainties, or user input.

2. Optimal and Adaptive Time-Varying Kinematic Controllers

Constrained Sequential Linear Quadratic (SLQ) for Mobile Manipulators

A major paradigm is the continuous-time, Constrained SLQ method, developed for high-degree-of-freedom (DoF) mobile manipulators with both nonholonomic and holonomic constraints (Giftthaler et al., 2017). The main workflow involves:

System Modeling: State $x(t)$ represents all robot coordinates (base and arm), with control $u(t)$ as their velocities. Unconstrained kinematics are $ẋ = u$ , enabling a perfect-velocity-tracking assumption at the kinematic level.
Constraint Incorporation: Nonholonomic constraints (e.g., rolling without slip) are enforced as $A_n(x) u = 0$ , while holonomic (task-space) constraints are encoded via $\phi(x) - x_{\mathrm{ref}}(t) = 0$ with possible velocity constraints $f$ 0.
LQ Problem: The continuous-time LQ objective, quadratic in $f$ 1, is solved iteratively by linearizing both dynamics and constraints about a time-varying nominal trajectory. Each SLQ iteration solves a constrained, time-varying LQR subproblem, leading to Riccati-type backward equations for the time-varying value function $f$ 2 and multipliers.
Feedback Law: The resulting optimal, time-varying kinematic feedback has the form

$f$ 3

with $f$ 4 and $f$ 5 computed to ensure first-order feasibility. Solution steps employ adaptive step-size ODE solvers for both forward and backward passes, yielding computational cost $f$ 6 per iteration (where $f$ 7 is adaptively chosen), and supporting real-time deployment at up to $f$ 8 Hz for complex systems (Giftthaler et al., 2017).

LTV–LQR and Model-Based Approaches

Controllers based on linear time-varying (LTV) models, most notably the LTV–LQR, address time–indexed trajectory tracking with explicit gain adaptation. In human–robot shared-control tasks, candidate grasps are precomputed as time-parameterized trajectories, and an LTV–LQR is synthesized around each. Gains are computed by integrating the backward Riccati equation with time-varying weights, and real-time selection and blending are executed according to the user’s inferred intent (Zito et al., 2019). This approach provides $f$ 9-tracking guarantees via a time-varying Lyapunov function.

In optimal kinematic tracking for manipulators, adaptive critic designs (e.g., Single Network Adaptive Critic, SNAC) approximate the solution of the time-varying Hamilton-Jacobi-Bellman equation via neural networks, supporting nonlinear, time-varying references (Menon et al., 2019).

3. Synthesis Under Constraints and Safety Guarantees

Control Barrier Functions for Time-Varying Safety

Time-varying kinematic control increasingly leverages control barrier functions (CBFs) for enforcing safety over moving, nonstatic regions. Uniformly time-varying CBFs require

$x_d(t)$ 0

where $x_d(t)$ 1 is a class- $x_d(t)$ 2 function (Wiltz et al., 2024). Efficiently, one constructs $x_d(t)$ 3-shiftable CBFs: starting from an invariant $x_d(t)$ 4, one certifies that $x_d(t)$ 5 (where $x_d(t)$ 6) remains a valid time-varying CBF, yielding forward-invariance of moving safe sets and supporting quadratic-program synthesis for safe tracking.

Hamilton–Jacobi Reachability and Tube-based Tracking

For systems with unknown, time-varying disturbances or constraints, safety-guaranteed time-varying tracking is achieved by offline solving the time-varying Hamilton-Jacobi–Isaacs variational inequality for the relative error model. The value function $x_d(t)$ 7 characterizes the minimal cost-to-go or maximal deviation, defining a tube $x_d(t)$ 8 within which the tracking error is certified invariant (Siriya et al., 2024). Online, a reference trajectory for a virtual or simplified planning model is (re)planned as constraints evolve, and the actual system is kept within tracked tubes via precomputed policies $x_d(t)$ 9. This framework admits extensions to $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 0-periodic dynamics via horizon mapping.

4. Formation, Swarm, and Multi-Agent Control with Time-Varying Objectives

Leader–Follower and Shape-Morphing Strategies

Time-varying leader–follower formation control mandates that the relative positions, distances, and bearings in a formation adapt according to explicit, possibly complex, time-dependent profiles (e.g., $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 1). Backstepping methodologies admit time-derivatives of these quantities, supporting globally asymptotically stable tracking of time-varying geometries (Nourizadeh et al., 2022). Fuzzy–adaptive augmentations enable online tuning of gains to mitigate transient actuator saturation and "velocity jump" artifacts.

Swarm Formation and Target Interception

Distance-based frameworks for multiple unicycle-type robots employ input transformations to achieve effective single-integrator-like equations, enabling time-varying flocking velocity and interception protocols. Here, only a subset of agents may have direct access to global time-varying references, while the rest reconstruct them via distributed, variable-structure observers. Stability of the entire formation is established via Lyapunov techniques and input-to-state stability interconnection arguments (Khaledyan et al., 2018).

Coverage Control Over Moving Domains

Time-varying multi-agent coverage is formulated as the minimization of locational cost over dynamic, possibly shape-changing domains, employing centralized or decentralized controllers that account for both local cell-coupling (via the Jacobian $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 2) and explicit feedforward terms $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 3 capturing domain motion and density evolution (Xu et al., 2019). Exponential convergence to moving centroidal Voronoi tessellations is achievable under analytic simplifications and scalable decentralized control.

5. Prescribed-Time and Nonholonomic Kinematic Control

Time-varying high-gain feedback laws achieve exact prescribed-time convergence for uncertain nonholonomic chained systems by smooth, time-varying coordinate transforms and parametric Lyapunov equations (Zhang et al., 2024). The key mechanism involves time-blowup scaling, where the feedback gain increases as $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 4, ensuring that all states reach zero at the designer-chosen $J = m(x(t_f)) + \int_{t_0}^{t_f} \ell(x(t), u(t), t)\,dt,$ 5 regardless of initial condition or triangular-structured uncertainty.

Nonholonomic mobile robots employing time-varying, oscillatory gradient-flow-based control exploit the Lie-algebraic structure to average out non-integrable constraints, enabling robust stabilization via periodic excitation (with engineered amplitude, frequency, and Lyapunov function shape) (Zuyev et al., 22 Feb 2026). Time-optimal velocity tracking for differential-drive robots in the presence of time-varying reference velocities uses hybrid, finite-switching logic to ensure minimal convergence time without chattering (Poonawala et al., 2017).

6. Adaptation, Learning, and Robustness to Uncertainty

Model-based reinforcement learning for time-varying dynamics uses Gaussian process priors and variation-budget analysis to handle non-stationarity across episodes (e.g., due to wear, payload change, unmodeled disturbances) (Iten et al., 2 Apr 2026). Recency-weighted buffers and adaptive planning maintain epistemic uncertainty and support dynamic-regret guarantees, ensuring continual adaptation in changing kinematic regimes. Learning mechanisms are tightly integrated with time-varying feedback policy design, supporting autonomous or semi-autonomous agents operating in dynamic real-world environments.

7. Teleoperation, Delay-Robustness, and Architectural Constraints

Teleoperation with closed-architecture robots under arbitrary bounded, time-varying delay employs dynamic feedback controllers combining local error, delayed peer state, and integral actions to guarantee delay-independent synchronization and infinite manipulability of degree one (Wang et al., 2021). Adaptive dynamic augmentation further compensates for unknown inner-loop PID gains, while input-output stability is ensured even as time-dependent delays alter the effective communication topology.

Time-varying kinematic control constitutes a mathematically and practically sophisticated domain embedding optimality, safety, learning, and coordination. Across real-time manipulator control, robust mobile robot planning, adaptive multi-agent coverage, and teleoperation under delay, the field is shaped by continuous methodological advances in feedback law design, constraint handling, and algorithmic scalability (Giftthaler et al., 2017, Zito et al., 2019, Khaledyan et al., 2018, Zuyev et al., 22 Feb 2026, Poonawala et al., 2017, Iten et al., 2 Apr 2026, Wang et al., 2021, Wiltz et al., 2024, Siriya et al., 2024, Zhang et al., 2024, Xu et al., 2019, Nourizadeh et al., 2022, Menon et al., 2019).