Active Localization of Unstable Systems with Coarse Information

Ege Yuceel, Daniel Liberzon, Sayan Mitra

This paper, recently published by researchers at the University of Illinois Urbana-Champaign, explores a fascinating corner of control theory: how to find yourself when you have almost no information and the world is trying to push you away.

TL;DR

The authors prove that an agent can perfectly localize its initial position using only a single bit of information (a "yes/no" proximity sensor) even when the system's dynamics are unstable. They introduce ActiveLocalize, an algorithm that uses instability to its advantage to exponentially shrink uncertainty about the agent's starting state.

What Problem Does This Paper Address?

In robotics, we usually assume a firehose of data: high-res cameras, LiDAR, and IMUs. But what happens in "minimalistic robotics"—think tiny, low-power sensors or agents operating in extreme environments?

The authors look at the "extreme sensing regime":

  1. Coarse Information: The agent only knows if it is within a certain radius of a landmark ($y=1$) or not ($y=0$). This is a single-bit, binary measurement.
  2. Unstable Dynamics: The system is naturally "repulsive." If left alone, the agent will drift away from the landmark exponentially fast.

The challenge is two-fold. First, how do you estimate where you started ($x_0$) when your only feedback is a series of 1s and 0s? Second, because the system is unstable, how do you prevent the agent from drifting into the "dark" (the $y=0$ zone) forever, where it receives no new information?

Methods

The authors propose a dual-strategy algorithm called ActiveLocalize.

  • Set-Based Estimation: When the agent receives a "1" (it's near the landmark), the algorithm treats this as a geometric constraint. By looking at the history of controls and measurements, it constructs ellipsoids in the state space. The intersection of these ellipsoids represents the set of possible initial states.
  • Spherical Voronoi Partitions (SVP): This is the "active" part of the localization. To ensure the agent doesn't get lost, the authors divide the space around the landmark into regions using SVPs.
  • Recovery Control Signal (RCS): If the agent loses the signal ($y=0$), it triggers a recovery sequence. Based on which Voronoi cell the agent's estimate is in, the algorithm applies a specific control sequence designed to steer the agent back into the sensing ball.

The "secret sauce" is the mathematical proof that instability actually makes states more distinguishable. In a stable system, two different starting points might produce the same string of 1s and 0s forever. In an unstable system, those trajectories diverge, eventually forcing different measurement sequences.

Datasets

The paper does not use standard real-world datasets. Instead, it evaluates the algorithm using randomly generated numerical simulations of discrete-time linear systems ($2 \times 2$ matrices).

  • Setup 1: Minimum absolute eigenvalue $\lambda_{min} = 1.014$.
  • Setup 2: Minimum absolute eigenvalue $\lambda_{min} = 1.010$. The authors ran 40 independent trials for each setup to validate the theoretical convergence bounds.

Key Insights and Ideas

  • Instability is a Feature, Not a Bug: Counter-intuitively, instability helps localization. Because trajectories diverge quickly, any "wrong" guess about the initial state will eventually produce a trajectory that misses the landmark when the true state hits it. This allows the algorithm to prune the set of possible initial states much faster than in a stable system.
  • Indistinguishable Sets Collapse: In previous work on stable systems, it was shown that you often can't shrink your uncertainty to a single point—you're left with a "diagonal" of possible states. This paper shows that under unstable dynamics, these sets collapse to singletons, meaning perfect localization is theoretically possible.
  • Guaranteed Re-entry: The authors prove that their Voronoi-based recovery strategy will always bring the agent back to the sensing region within a finite number of steps, provided the landmark uncertainty isn't too large relative to the system's instability.

Key Results

  • Exponential Contraction: The primary result is a proof that the diameter of the initial-state uncertainty set shrinks exponentially: $diam(\hat{X}_0[k]) \leq C \cdot a^k$, where $0 < a < 1$.
  • Frequent Hits: The algorithm guarantees that the agent will visit the sensing ball infinitely often, with no more than a fixed number of steps ($N\bar{n}$) between positive measurements.
  • Numerical Validation: Simulations confirmed that the uncertainty diameter drops to near-zero within a few hundred iterations, consistently staying below the predicted theoretical upper bounds.

Limitations and Future Work

  • Linear Systems Only: The current proofs rely on linear dynamics ($x_{k+1} = Ax + Bu$). Real-world robots are often nonlinear.
  • No Noise: The model assumes "perfect" binary sensing and no process noise. In reality, sensor jitter or wind (process noise) could make the "1" and "0" boundaries fuzzy.
  • Single Landmark Focus: While the authors sketch out how to handle multiple landmarks, the core theory is optimized for a single stationary point.
  • Future Directions: The authors suggest extending this to hybrid systems (systems that change behavior abruptly) and more complex "coarse" sensors, like those that only detect line-based features or specific image keyframes.

Reference: Yuceel, E., Liberzon, D., & Mitra, S. (2026). Active Localization of Unstable Systems with Coarse Information. arXiv:2602.06191v1 [cs.RO].

Summary updated

Abstract

We study localization and control for unstable systems under coarse, single-bit sensing. Motivated by understanding the fundamental limitations imposed by such minimal feedback, we identify sufficient conditions under which the initial state can be recovered despite instability and extremely sparse measurements. Building on these conditions, we develop an active localization algorithm that integrates a set-based estimator with a control strategy derived from Voronoi partitions, which provably estimates the initial state while ensuring the agent remains in informative regions. Under the derived conditions, the proposed approach guarantees exponential contraction of the initial-state uncertainty, and the result is further supported by numerical experiments. These findings can offer theoretical insight into localization in robotics, where sensing is often limited to coarse abstractions such as keyframes, segmentations, or line-based features.
BibTeX
@misc{yuceel2026active,
  author = {Ege Yuceel and Daniel Liberzon and Sayan Mitra},
  title = {{Active Localization of Unstable Systems with Coarse Information}},
  year = {2026},
  eprint = {2602.06191},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2602.06191},
}
Figure