An elevator could change its stopping plan halfway through a ride without the change itself producing a lurch. That is the striking promise of profile-anchored control with exact rejection, or PACER, a control method developed by Christian Fredrik Sætre and Morten Dinhoff Pedersen. In their theory, a controller can recover a prescribed motion after a disturbance has upset it, then replace that motion profile mid-course and follow the new one from the current state—without a switching transient or restarting its estimate of the disturbance.
The result is not a report of elevators already using this method. The paper proves properties of a mathematical control design and illustrates it in a simulated elevator. Its broader idea matters well beyond lifts: in trains, robots, or other machines, the desired route through speed and deceleration can become part of the feedback controller itself, rather than a reference the machine is merely asked to chase.
A motion plan is often written against time: be here at this moment, then there at the next. But if a load changes, a door delays departure, or an external force disrupts motion, a clock-based plan can quickly become stale. A position-based profile answers a different question: given how far remains to the target, what velocity should the machine have now? That makes a profile more like a reusable rule for approaching a destination than a schedule tied to a particular start time.
The Science
The authors study a one-dimensional system moving toward a target. Let position relative to the target be represented by a variable, and let velocity be specified as a function of that position. The profile points toward the target: as the machine gets closer, its prescribed velocity guides it toward rest. A profile can encode a speed cap or a desired pattern of acceleration and deceleration.
The key conceptual move is to use the square of the velocity profile. If the profile is called , the controller uses . Along the desired motion, the required nominal acceleration is the derivative , rather than a term that directly depends on the slope . This matters because some useful profiles have a sharp or ill-behaved slope near the target, even though their squared profile remains smooth enough to use in control.
A controller needs more than a nominal plan. Real mechanisms face unknown loads and disturbances, and the same control input may have an uncertain effect. The paper first develops a static feedback law for a second-order system: one in which acceleration responds to the input. The law combines the profile’s nominal acceleration with feedback that pulls the actual speed toward the speed prescribed at the current position. Under stated assumptions—including an unknown but positive input gain with a known lower bound and a bounded perturbation—the authors prove global asymptotic stability. In plain language, the system eventually settles at the target from any starting state, with the uncertainty constrained as specified.
They also establish finite-time stability for a class of profiles with a particular local scaling property, called homogeneity. Rather than merely approaching the target ever more closely, the idealized model reaches it in finite time. The paper gives familiar controllers as special cases, including proportional-derivative control and finite-time sliding-mode designs. That connection is useful: PACER is not a wholly separate universe of control, but a framework that can include known feedback structures while making the desired path through velocity explicit.
The static controller has a limitation. An unknown load can keep pushing the machine off the nominal profile, even when the profile itself is well designed. To address persistent disturbances, the authors construct a dynamic law for a first-order system. Dynamic means the controller has its own evolving internal state, which acts like integral action: it can build an estimate of the persistent force or bias rather than trying to cancel it with instantaneous feedback alone.
This part connects the framework to PI control and the super-twisting algorithm, a robust sliding-mode method designed to handle certain disturbances without relying on an exact disturbance value. The paper’s formal guarantee covers disturbances whose rate of change is bounded. The controller does not need the disturbance to be zero; under the stated conditions, it can compensate for its effect.
PACER combines the static and dynamic constructions. Its central result is that the actual system can recover the nominal profile motion exactly after a finite time, under an unknown input gain and time-varying loads whose rate is bounded. “Exactly” refers to the modeled dynamics after the recovery time, not to perfect performance in every physical elevator. The result is semiglobal: the guarantee applies over a chosen bounded operating region, rather than uniformly across every possible state and machine configuration.
The architecture is designed to make replanning natural. Because the desired speed depends on position rather than elapsed time, a new profile can be defined through the current position and velocity. The paper proves that the replacement profile can take effect from the switching point without a transient, and without resetting the internal disturbance estimate. This is a theoretical switching property, not a claim that any arbitrary profile can be switched in under any actuator condition.
What They Found
The authors’ results are primarily mathematical guarantees, not measured performance statistics. The paper provides no real-world dataset from which to plot elevator ride times, energy use, passenger comfort, or failure rates. So the most useful “numbers” here are the quantified conditions and the distinctions among the claims—not invented bar charts that imply a physical trial.
For the second-order feedback, the unknown input gain may vary over a range, but it must be at least a known positive lower bound. The allowable disturbance is constrained relative to the corrective feedback. Under those conditions, the origin is uniformly globally asymptotically stable. With additional local homogeneity conditions, the theorem strengthens this to finite-time stability.
For the dynamic disturbance-rejection component, the disturbance itself may change over time, but its rate of change has to remain within a specified bound. Under that condition, the closed loop is globally asymptotically stable; with additional profile behavior near the target, the paper establishes finite-time stability. These are conditional guarantees. The formulas do not tell us how much noise a particular commercial elevator controller can tolerate until someone maps the model assumptions onto that hardware.
A further result concerns staying inside a velocity envelope. For some settings and sign conditions, the set of states that are moving toward the target and not exceeding the profile is forward invariant: once the system begins within the envelope, it stays there. The authors emphasize that this does not follow automatically for every unknown gain and every profile. In particular, gain mismatch can cause an overshoot of the profile where the desired motion is accelerating. The dynamic extension is intended to address that mismatch.
The planning result links this control design to optimization. The profile can be replanned subject to speed, acceleration and jerk limits, where jerk is the rate of change of acceleration. Those constraints can be written as a convex program in , the squared-profile quantity that also determines nominal control. Convexity is valuable because it gives planners a well-structured optimization problem rather than a search through arbitrary motion curves. The supplied paper excerpt does not include numerical elevator simulation results or plotted data, so there are no honest empirical charts to reproduce.
In the simulated elevator example, the point is to illustrate how profile shaping, disturbance rejection and switching fit together. The example is a proof of concept in simulation, not a field study comparing PACER against existing lift controllers. Nor does the paper report passenger comfort scores, energy savings, travel-time reductions, or safety incidents avoided. Those outcomes are plausible reasons to care about the method, but they remain questions for future engineering work.
Why This Changes Things
A time-stamped trajectory can be brittle when the world does not follow its schedule. Suppose a train pauses longer than planned, or a robot encounters an object that changes its motion. A controller that insists on catching up with the original clock may demand an aggressive correction. A position-based velocity profile instead says how to move from wherever the machine is now, in relation to the destination. It can resume the approach without pretending that the delay never happened.
That distinction also changes how planning and control fit together. In many systems, a planner selects a trajectory and a separate feedback loop tries to track it. If tracking fails or conditions change, another layer may be needed to enforce limits or generate a new trajectory. PACER’s design puts the profile directly inside the controller, while its optimization formulation uses the same squared profile that defines nominal input. The potential benefit is a tighter link between what the plan says and what the controller can execute.
The transient-free switch is especially relevant where plans need to change while a machine is moving. A lift might need to respond to a changed destination or loading estimate; a robot may need to revise its approach as a task evolves. If the new profile passes through the current state, the theory says the controller can switch without injecting an additional transient. Keeping the disturbance estimate also avoids throwing away information the controller has already accumulated.
But a mathematical guarantee is only as useful as the match between its assumptions and the machine. The paper assumes a scalar motion variable, a constant but uncertain input gain with a known lower bound, and disturbances subject to specified restrictions. Real systems have sensor noise, actuator saturation, delays, friction that can change abruptly, coupled motion in several directions, and safety rules that may not fit a single profile. The convex planning formulation addresses stated speed, acceleration and jerk limits, but the excerpt does not establish how every practical constraint can be incorporated.
There is also an important difference between keeping speed within a profile and recovering the nominal motion eventually. The paper’s basic static law guarantees convergence but does not by itself make the profile an invariant boundary under uncertainty. Stronger conditions are needed for the envelope result, and the paper is explicit about this. For designers, that is not a weakness to hide; it is a map of which guarantee applies to which operating assumptions.
What's Next
The next step is to test the assumptions against hardware. An elevator experiment could measure how profile recovery behaves under changing payloads, friction, sensor noise and actuator limits, then compare comfort and travel time with a conventional controller. In robotics, trials could ask whether replanning during contact or obstacle avoidance stays smooth when the model is imperfect. Such tests would reveal not just whether the mathematics works, but how much engineering is needed to make it robust in the real world.
Further work could extend the approach beyond a single motion coordinate and examine practical issues such as sampled sensors, computation delay and saturated actuators. The paper’s convex formulation offers a promising route for generating profiles, but an operational planner must also account for discrete events, uncertainty and competing goals. It will matter whether a newly planned profile can be found quickly enough when conditions change, and whether its limits remain safe when the estimated machine parameters are wrong.
For now, PACER is best understood as a control framework with a sharp central insight: prescribe motion as velocity versus position, use the squared profile to connect planning with nominal control, and add dynamic feedback to reject persistent disturbances. The paper proves that this combination can recover a desired motion and switch profiles cleanly within a defined model. If hardware trials bear out those properties, machines may gain a more graceful way to adapt—not by chasing an obsolete timetable, but by reshaping the approach from where they actually are.