Delay-Certified Readiness Barriers for Overactuated Multirotor Allocation Without Lipschitz Constants

Giuseppe SilanoarXiv 2026 · arXiv

TL;DR

This paper introduces a "Lipschitz-free" delay certificate for overactuated multirotors to ensure they always maintain enough control authority (readiness) to reject disturbances. By exploiting the monotonic nature of rotor dynamics, the authors provide a computationally efficient way to guarantee safety even with significant communication delays and timing jitter, outperforming traditional methods that rely on conservative mathematical bounds.

What Problem Does This Paper Address?

Overactuated multirotors (like hexarotors) have more motors than they have degrees of freedom to control. This redundancy means there are many ways to distribute the workload among rotors to achieve a specific "wrench" (the combined force and torque on the body). While most allocators focus on minimizing energy, they often ignore readiness: the vehicle's ability to change its wrench rapidly to fight off a sudden gust of wind.

Previous research used Control Barrier Functions (CBFs) to keep this readiness above a safe "floor." However, these guarantees were designed for continuous time. In the real world, flight controllers send sampled commands through Electronic Speed Controllers (ESCs) that introduce varying delays. Standard ways to make CBFs robust to delay rely on "Lipschitz constants"—global bounds on how fast a system can change—which are often so conservative that they only certify delays of less than one millisecond. In practice, delays are often much longer, leaving a dangerous gap between what is mathematically "certified" and how the drone actually performs.

Methods

The authors bridge this gap by moving away from generic bounds and looking at the specific physics of the problem. They leverage three structural facts:

  1. Monotonicity: Each rotor responds to torque independently and predictably. If you apply more torque, the rotor speed increases; it never "surprises" the system.
  2. Closed-form Flow: The speed of a rotor under constant torque can be calculated exactly without complex integration.
  3. Unimodality of Readiness: For any single rotor, its contribution to the vehicle's overall readiness peaks at a "sweet spot" speed and drops off if it goes too fast or too slow.

By combining these, the authors can calculate a "box" of all possible rotor speeds that could occur during a delay window. Because of the unimodality, the absolute minimum readiness over that entire box is guaranteed to be at one of the corners.

The certificate process: (a) command envelopes, (b) reachable speed boxes, and (c) the unimodal readiness weight.

They developed Algorithm 1, which predicts the future state, checks if a candidate command keeps readiness above the floor for the entire delay window, and—if not—blends the command with a "sweet-spot hold torque" that acts as a certified safety net.

Datasets

The study is evaluated using high-fidelity simulations of a coplanar hexarotor model. The primary benchmark is a "roll-moment reversal" mission, where the drone must rapidly flip its roll torque. The authors tested this across:

  • Certifiable missions: Demands stay within the vehicle's design envelope.
  • Readiness-priority missions: Demands exceed the envelope, forcing the filter to trade off tracking accuracy for safety.
  • Varying Delays: 2 ms to 40 ms of nominal delay with up to ±2 ms of random jitter.

Key Insights and Ideas

  • Structure Trumps Generality: By using the specific monotonic structure of rotor dynamics, the authors avoided the "conservatism tax" of Lipschitz constants. This allowed them to certify delays 10–40 times longer than previous methods.
  • The "Sweet Spot" Fallback: The discovery that readiness is unimodal led to a "hold torque" strategy. If a new command is risky, the system can fall back to a torque that drives rotors toward their most "ready" state, ensuring recursive feasibility (there's always a safe move available).
  • Exact Box Minimums: Finding the worst-case readiness over a range of speeds usually requires a complex search. Here, it costs just a few scalar evaluations because the minimum is always at a known corner of the reachable speed box.

Key Results

  • Zero Violations: In simulations with delays up to 40 ms and jitter, the proposed filter never allowed the readiness to drop below the floor. In contrast, standard filters and uncertified predictors failed frequently at delays beyond 5–10 ms.
  • Efficiency: The safety check is incredibly fast, taking only about 9.6 microseconds on a standard CPU. This makes it highly practical for real-time flight controllers.
  • Robustness: The "Robust" version of the certificate (Algorithm 1-R) maintained safety even with ±10% errors in aerodynamic drag coefficients and speed measurement noise.
Fig. 2: Nominal-delay sweep at jitter ±2 ms (∆d = 4 ms), α = 5, synchronous switching, five jitter realizations per point; c = 0.75, (a) β = 1.5, ¯ℓ= −11.061, (b), (c) β = 2, ¯ℓ= −11.076. Pred-Lip-Loc was not run in this sweep. (a), (b): worst mint h over the realizations (below zero: floor violated); (c): median RMS wrench error. Algorithm 1 stays above the floor at every delay, while the RB filter and SD-Lip fail at long delays; in (a), Algorithm 1 and Pred coincide.

Limitations and Future Work

  • Model Dependence: The certificate relies on the specific rotor model (first-order dynamics with quadratic drag). While common, it might need adjustment for more exotic propulsion systems.
  • Task-Agnostic Fallback: The current fallback strategy (the hold torque) prioritizes readiness over the mission goal. This can cause the drone to deviate from its path (e.g., losing altitude) to stay "safe." The authors suggest future work on "wrench-preserving" backups that maintain the flight path while recovering readiness.
  • Hardware Validation: While the simulation results are robust, the next logical step is testing on physical hardware with measured ESC delays.

Reference: G. Silano, "Delay-Certified Readiness Barriers for Overactuated Multirotor Allocation Without Lipschitz Constants," arXiv:2611.03490v1, 2026.

Summary updated

Abstract

Overactuated multirotors can produce the same wrench with many rotor-speed combinations, some of which leave the vehicle unable to change its wrench quickly, e.g., to reject a gust. Readiness barriers keep this ability, the readiness, above a floor, but only in continuous time; with sampled commands and varying speed-controller delays, the available robustness bound covers less than one millisecond. We remove this limit by exploiting the structure of the problem: each rotor responds monotonically and in closed form to its own torque, so the speeds reachable under uncertain switching times lie in a cheaply computed box, and the lowest readiness over that box sits at a corner known in advance. This yields a Lipschitz-free test that a command keeps the readiness above its floor until the next sample, plus a fallback torque that keeps the vehicle on a certified plan. In hexarotor simulations with 2--40~ms delays and $\pm2$~ms jitter, the filter never violated the floor. At $10\pm2$~ms, it matched a delay-compensating predictor within the design envelope and cost 2--30\% more tracking error beyond it, where the predictor alone violated the floor in 2 to 16 of 16 missions.
BibTeX
@misc{silano2026delay,
  author = {Giuseppe Silano},
  title = {{Delay-Certified Readiness Barriers for Overactuated Multirotor Allocation Without Lipschitz Constants}},
  year = {2026},
  eprint = {2611.03490},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2611.03490},
}
Figure