Prescribed Performance Control

• Prescribed Performance Control (PPC) is a control design paradigm that bounds tracking errors within time-varying envelopes, encoding transient behavior and final accuracy directly.
• It employs state transformations such as logarithmic barrier maps and error normalization to convert constrained tracking problems into stabilizable forms for various systems.
• PPC integrates with architectures like backstepping, dynamic surface, and sliding-mode controls, while adapting to constraints such as actuator limits, delays, and saturation.

Prescribed Performance Control (PPC) is a control design paradigm that constrains tracking errors to evolve within designer-specified time-varying bounds, often described as performance envelopes or funnels, so that transient and steady-state requirements such as maximum overshoot, convergence rate, and terminal accuracy are encoded directly in the control problem. Across the formulations reported for strict-feedback nonlinear systems, underactuated aerial vehicles, spacecraft attitude dynamics, heterogeneous multi-agent systems, underwater gliders, sampled-data systems, and systems with delays, PPC is used to guarantee inequalities of the form $|e(t)|<\rho(t)$ or their asymmetric counterparts, while the controller is designed on transformed or normalized errors so that the original constrained error remains inside the prescribed set for all relevant times (Lapandić et al., 2022, Mishra et al., 2022, Shivam et al., 16 Feb 2026).

1. Core formulation and prescribed-performance envelopes

At its most standard, PPC specifies a scalar tracking-error constraint by prescribing a positive performance function and requiring the error to remain inside it: $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$ or, more generally, an asymmetric bound such as

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$

The function $\rho(t)$ or $P_i(t)$ is smooth, positive, and usually decreasing, thereby encoding initial admissible error, transient contraction, and a nonzero terminal bound (Lapandić et al., 2022, Yang et al., 23 Dec 2025).

A common exponential choice is

$$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$

with $\rho_0>\rho_\infty>0$ and $\lambda>0$. This directly specifies the initial envelope, the contraction rate, and the steady-state accuracy. Closely related exponential envelopes appear in spacecraft pointing, unknown strict-feedback systems with input constraints, and hybrid-gain sliding-mode designs (Shivam et al., 16 Feb 2026, Mishra et al., 2022, Lei et al., 2022).

The same idea is also used in funnel-control language. For reentry vehicles, the admissible set is written as

$$\Gamma_{\varphi_i}:=\{(t,e_i)\mid \varphi_i(t)|e_i|<1\},$$

with $\bar\rho_i(t)=1/\varphi_i(t)$ interpreted as the funnel boundary. In that formulation, prescribed performance means that the tracking error remains within a time-varying funnel whose width can itself be redesigned during maneuver phases (Guo et al., 2023).

The envelope need not be symmetric or fixed-shape. In heterogeneous UAV–UGV formation control, the position error is constrained by

$-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$0

where the time-varying bounds $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$1 and $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$2 adapt to actuator saturation. In underwater-glider path following, different left and right margins are used through $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$3 and $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$4, so that PPC explicitly accommodates asymmetric overshoot allowances (Zhang et al., 10 Apr 2025, Yang et al., 23 Dec 2025).

These formulations indicate a unifying viewpoint: PPC does not primarily prescribe a controller structure, but rather a state-dependent inequality that must remain true throughout closed-loop evolution.

2. Error normalization, transformations, and performance-function design

The basic PPC mechanism introduces a normalized error, typically

$-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$5

or one of its asymmetric generalizations, and then maps the constrained interval for $-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$6 into an unconstrained variable. A canonical choice is the logarithmic barrier-type mapping

$-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$7

which appears in observer and formation-error transformations for heterogeneous multi-agent systems and in standard PPC constructions for underactuated quadrotors (Zhang et al., 10 Apr 2025, Lapandić et al., 2022).

Several works depart from this logarithmic form. One line uses barrier Lyapunov functions directly on the normalized error instead of an explicit logarithmic state transformation. For spacecraft reduced-attitude pointing, the barrier term

$-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$8

is defined on

$-\rho(t)<e(t)<\rho(t),\qquad t\ge 0,$9

so that the PPC layer is combined with obstacle-avoidance logic without relying on the standard PPC logarithmic map (Lei et al., 2022).

Another line embeds the constraint through a smooth inverse-error-function representation: $-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$0 Because $-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$1 for finite arguments, boundedness of $-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$2 implies $-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$3. This construction is central to the prescribed-performance-aware hybrid-gain finite-time sliding-mode controller (Shivam et al., 16 Feb 2026).

Funnel-control variants use barrier factors directly in the feedback gain. For reentry vehicles,

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$4

so the gain diverges as the normalized error approaches the funnel boundary, preventing boundary crossing. That paper further replaces the usual monotone funnel by a time-triggered non-monotonic boundary built from exponential and cubic segments, allowing temporary widening during planned maneuvers (Guo et al., 2023).

A distinct singularity-avoidance route is the shear mapping-based error transformation

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$5

introduced for perturbed high-order systems. The purpose is global non-singularity: even if the normalized error temporarily leaves the conventional admissible interval because of abrupt reference changes, the transformed variable remains well defined (Liu, 19 Aug 2025).

Performance functions themselves also vary substantially across the literature. In addition to the exponential form, reported designs include:

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$6

for multi-agent observer and formation errors, which permits arbitrarily large initial errors and then enforces a prescribed steady-state bound after time $-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$7 (Zhang et al., 10 Apr 2025);

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$8

for finite-time performance in underwater gliders, where the bound reaches its final value at a preset time and then stays constant (Yang et al., 23 Dec 2025);

and a freezing performance-rate function for prescribed-time PPC,

$-\delta_{L,i}P_i(t)<e_i(t)<\delta_{R,i}P_i(t).$9

which is used both as a performance function and as a rate function, with $\rho(t)$0 to reduce initial control effort (Golestani et al., 21 Jan 2026).

Taken together, these designs suggest that PPC is better understood as a family of envelope-generation and constraint-embedding mechanisms than as a single normalized-error formula.

3. Controller architectures built around PPC

PPC is commonly embedded in backstepping, dynamic surface control, sliding-mode, and gradient-based architectures. In underactuated quadrotor tracking, a hierarchical PPC structure is used: position PPC generates a reference velocity, velocity PPC generates thrust and tilt-map commands, attitude PPC tracks tilt-map and yaw, and angular-rate PPC closes the inner loop, all without using model parameters or disturbance estimates (Lapandić et al., 2022).

Backstepping remains one of the most common realizations. For spacecraft reduced-attitude pointing under forbidden directions, the first-layer error is the normalized pointing error, the virtual control is a desired angular velocity that blends PPC attraction and artificial-potential-field terms, and the second layer uses the backstepping error

$\rho(t)$1

with torque injection along both the PPC and APF gradient directions (Lei et al., 2022).

Dynamic surface control is used to avoid the repeated differentiation of virtual controls. In practical prescribed-time PPC with asymptotic convergence, the transformed first-layer error is

$\rho(t)$2

higher-order errors are defined recursively, and nonlinear filters

$\rho(t)$3

replace direct differentiation of virtual controls (Golestani et al., 21 Jan 2026).

Sliding-mode-based PPC designs are also prominent. In heterogeneous UAV–UGV systems, the transformed formation error is used to define a sliding surface

$\rho(t)$4

and the distributed controller is synthesized so that the surface dynamics reduce to

$\rho(t)$5

thereby guaranteeing prescribed bounds on formation and observer errors despite communication failures and actuator saturation (Zhang et al., 10 Apr 2025).

The prescribed-performance-aware hybrid-gain finite-time sliding-mode controller combines PPC with a hybrid gain law. In the first-order case,

$\rho(t)$6

and in the second-order case the sliding variable is

$\rho(t)$7

The hybrid gain is piecewise, with outer and inner regions, so that finite-time convergence and bounded control effort are obtained simultaneously (Shivam et al., 16 Feb 2026).

Other reported architectures include fixed-time sliding-mode disturbance observers paired with PPC for underwater gliders (Yang et al., 23 Dec 2025), I&I adaptive composite controllers for reduced-attitude spacecraft alignment (Lei et al., 2023), approximation-free nested tangent–arctangent mappings for unknown strict-feedback systems with prescribed input constraints (Mishra et al., 2022), and derivative-free sample-and-hold output feedback for relative-degree-two systems, where the sample-based proxy

$\rho(t)$8

replaces any direct use of $\rho(t)$9 (Lanza, 2024).

4. Constraint handling, compatibility management, and robustness mechanisms

A central theme in recent PPC work is that prescribed error envelopes may conflict with actuator limits, safety constraints, delays, or communication failures. Several papers therefore replace fixed-envelope PPC by compatibility-aware or correction-based designs.

For unknown nonlinear strict-feedback systems with prescribed input constraints, PPC and actuator limits are treated jointly through a nested barrier–saturation mapping: $P_i(t)$0 leading to a controller that satisfies both $P_i(t)$1 and $P_i(t)$2, provided an explicit feasibility condition is met (Mishra et al., 2022).

Input saturation motivates adaptive envelope scheduling in other formulations. In self-adjusting PPC for nonlinear systems with saturation, the decay rate $P_i(t)$3 of the performance function is increased when the error remains inside a performance index function and decreased when the error approaches the outer prescribed envelope. The reported effect is that the controller can reduce decay rates when necessary to avoid violation of the PFs and increase them when there is remaining control capacity (Shao et al., 2024).

Spacecraft pointing under forbidden directions introduces another kind of incompatibility: rapid prescribed convergence can conflict with obstacle avoidance. One solution is a switched prescribed performance function with a freeze mechanism,

$P_i(t)$4

so that when $P_i(t)$5,

$P_i(t)$6

The normalized PPC error is then frozen while the artificial potential field takes precedence (Lei et al., 2022). Closely related spacecraft attitude work describes this as Compatible Performance Control, using contradiction detection based on a Zeroing Barrier Function and a projection-operator-governed envelope-modification signal (Lei et al., 2023). Reduced-attitude boresight alignment extends the same principle through a Switched Prescribed Performance Function that monitors forbidden cones, angular-velocity limits, and the PPC state itself (Lei et al., 2023).

Delays and communication failures introduce further modifications. For higher-order uncertain nonlinear systems with state-measurement and input delays, the normalized errors are built from delayed measurements together with a delay-dependent correction chain: $P_i(t)$7 The correction vanishes in the zero-delay limit, and the controller reduces to the nominal PPC envelope (Berger et al., 10 Sep 2025). In distributed multi-agent control under faulty directed graphs, prescribed bounds are retained by combining PPC transformations with a distributed observer based on the faulty Laplacian $P_i(t)$8 (Zhang et al., 10 Apr 2025).

Sampled-data settings require another adaptation. In derivative-free sample-and-hold PPC, only $P_i(t)$9 and $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$0 are used, and a sufficient uniform sampling-rate condition is derived so that the funnel inequality $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$1 remains valid globally under zero-order hold (Lanza, 2024).

These constructions show that PPC does not automatically solve compatibility problems; rather, recent work increasingly builds explicit contradiction detection, envelope relaxation, freezing, adaptation, or correction mechanisms around the performance function itself.

5. Domains of application and task formulations

The reported applications are unusually broad. In spacecraft control, PPC has been used for full attitude tracking with prescribed performance and singularity avoidance (Lei et al., 2022), reduced-attitude boresight alignment under forbidden directions (Lei et al., 2022), adaptive reduced-attitude control under safety constraints and uncertainty (Lei et al., 2023), and attitude control with angular-velocity limitation, input saturation, and time-varying parameter uncertainty through Compatible Performance Control (Lei et al., 2023).

In aerial robotics, PPC has been adapted to underactuated quadrotors by introducing virtual references that compensate for the thrust-direction coupling and by structuring the controller as nested PPC loops for position, velocity, attitude, and angular velocity (Lapandić et al., 2022). In atmospheric reentry, PPC appears in a funnel-control formulation with a time-triggered non-monotonic funnel that widens during rapid trajectory changes and then contracts again (Guo et al., 2023).

Networked and heterogeneous systems form another major application class. In directed UAV–UGV multi-agent systems, PPC is used simultaneously for leader-state observation and formation tracking, with variable prescribed performance boundaries that react to actuator saturation and with explicit handling of communication link failures (Zhang et al., 10 Apr 2025). For uncertain higher-order systems with delays, PPC is extended to measurement-delay and input-delay channels through delay-dependent envelope correction (Berger et al., 10 Sep 2025). In derivative-free sampled-data output tracking, the same prescribed-performance objective is retained despite zero-order hold and the absence of derivative measurements (Lanza, 2024).

Marine applications include fixed-time prescribed-performance path following of underwater gliders. There PPC is integrated with an iLOS guidance law and a fixed-time sliding-mode disturbance observer, with the finite-time performance function reaching its terminal value at a preset time and then remaining constant (Yang et al., 23 Dec 2025).

Formal-task control provides a different interpretation of PPC. For signal temporal logic specifications, PPC is applied directly to the robustness of predicates, so that a funnel on $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$2 is leveraged to satisfy formulas such as $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$3 and $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$4 (Lindemann et al., 2017). A reinforcement-learning extension uses a PPC base law to guide exploration in PI$$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$5, combining

$$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$6

with a robustness-penalized objective to improve task-satisfaction learning (Varnai et al., 2019).

This breadth suggests that PPC is not domain-specific. What changes across domains is the envelope definition, the transformation, and the mechanism used to reconcile performance with physical or logical constraints.

6. Guarantees, trade-offs, and recurring limitations

The strongest attraction of PPC is that the control objective is stated as an inequality on the original error. Depending on the architecture, the reported guarantees include funnel invariance $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$7 for all $$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$8, no-overshoot with respect to the prescribed envelope, terminal accuracy such as

$$\rho(t)=(\rho_0-\rho_\infty)e^{-\lambda t}+\rho_\infty,$$9

practical asymptotic stability, fixed-time convergence to residual sets, and prescribed-time entry into a desired region followed by asymptotic convergence (Lei et al., 2022, Shivam et al., 16 Feb 2026, Golestani et al., 21 Jan 2026).

At the same time, the literature repeatedly emphasizes trade-offs. Tight envelopes, small steady-state bounds, or large contraction rates improve nominal precision but can sharply increase control effort, aggravate saturation, and induce over-control near the boundary (Lei et al., 2022, Shao et al., 2024). For simultaneous PPC and input constraints, arbitrary prescription is explicitly ruled out; a feasibility condition is required (Mishra et al., 2022). Delay-dependent correction mechanisms can be conservative because the admissible-delay bound enters the stability margin directly (Berger et al., 10 Sep 2025). In prescribed-time and finite-time variants, feasibility is still constrained by available control authority and disturbance bounds (Golestani et al., 21 Jan 2026, Shivam et al., 16 Feb 2026).

Several common misconceptions are directly contradicted by the reported results. PPC is not limited to one transformation; logarithmic barriers, BLFs, erf maps, funnel gains, and shear mappings all appear in the literature (Zhang et al., 10 Apr 2025, Lei et al., 2022, Shivam et al., 16 Feb 2026, Liu, 19 Aug 2025). PPC also does not automatically resolve conflicts with safety or actuator limits; obstacle avoidance, angular-velocity constraints, saturation, and delay often require switched envelopes, freeze mechanisms, variable boundaries, or compatibility analysis (Lei et al., 2022, Lei et al., 2023, Shao et al., 2024, Berger et al., 10 Sep 2025).

The limitations are equally consistent across papers. Many designs assume known control direction or positive input gain (Shivam et al., 16 Feb 2026), measurable full state or disturbance terms (Guo et al., 2023), bounded disturbances and known inertia structure (Lei et al., 2022), or stable zero dynamics and sufficient sampling rates (Lanza, 2024). Some formulations acknowledge conservative conditions, such as the trade-off inequality in maneuvering reentry tracking (Guo et al., 2023) or the delay-dependent smallness condition in higher-order systems with delays (Berger et al., 10 Sep 2025). Others explicitly note scope restrictions: matched disturbances for the hybrid-gain PPC–SMC design (Shivam et al., 16 Feb 2026), or the absence of a separate convex-hull proof in heterogeneous containment despite the formation construction (Zhang et al., 10 Apr 2025).

Overall, the published developments portray PPC as an envelope-centric methodology for control design. Its enduring technical content lies in specifying a performance corridor first, transforming the constrained problem into a stabilizable one second, and then modifying the corridor itself whenever safety, saturation, delays, sampling, or formal-task semantics make a fixed funnel incompatible with the actual closed-loop physics.