Optimal Byzantine Resilient Convergence in Asynchronous Robot Networks

We propose the first deterministic algorithm that tolerates up to $f$ byzantine faults in $3f+1$-sized networks and performs in the asynchronous CORDA model. Our solution matches the previously established lower bound for the semi-synchronous ATOM model on the number of tolerated Byzantine robots. Our algorithm works under bounded scheduling assumptions for oblivious robots moving in a uni-dimensional space.

💡 Research Summary

The paper addresses one of the most challenging problems in distributed robotics: achieving convergence of a swarm of mobile robots when up to f of them may behave arbitrarily (Byzantine faults) in an asynchronous environment. While prior work has established tight bounds for the semi‑synchronous ATOM model (namely that at least 3f + 1 robots are required to tolerate f Byzantine agents), no deterministic solution was known for the fully asynchronous CORDA model, where the Look‑Compute‑Move cycles of different robots can interleave arbitrarily.

The authors first formalize the asynchronous CORDA model with oblivious robots moving on a one‑dimensional line. They introduce a “bounded scheduling” assumption: every correct robot is guaranteed to be activated at least once within any sufficiently large but finite time window. This assumption is realistic in practice because it can be enforced by timeout mechanisms, periodic beacons, or fairness guarantees of the underlying scheduler.

The core contribution is a deterministic algorithm that works under the bounded‑scheduling assumption and tolerates up to f Byzantine robots provided the total number of robots n ≥ 3f + 1. The algorithm proceeds in repeated rounds, each consisting of three phases:

1. Observation – each robot obtains the current positions of all robots (including Byzantine ones, which may report arbitrary coordinates).
2. Computation – the robot computes the median of the observed positions. Because at least 2f + 1 of the n observations come from correct robots, the median is guaranteed to lie inside the convex hull of the correct robots’ positions, regardless of the Byzantine reports.
3. Movement – the robot moves toward the median but travels at most half the distance between its current location and the median. This “half‑step” rule ensures that the diameter (the maximum distance between any two correct robots) shrinks by at least a factor of two in each round where the robot is activated.

The safety proof hinges on the median property: with n ≥ 3f + 1, the set of correct robots forms a strict majority, so any median of the full multiset cannot be pulled outside the interval spanned by the correct robots. Consequently, the target point always remains within the safe region. The liveness proof shows that, under bounded fairness, each correct robot is activated infinitely often, and each activation reduces the diameter of the correct set by at least a factor of two. After O(log D/ε) activations (where D is the initial diameter and ε the desired precision), all correct robots are within ε of each other, i.e., convergence is achieved.

Key technical insights include:

• Bounded Scheduling as a Minimal Asynchrony Model – By requiring only that no correct robot can be starved indefinitely, the algorithm tolerates the full asynchrony of CORDA while still guaranteeing progress.
• Deterministic Median Selection without Memory – The algorithm does not rely on any persistent state; each robot recomputes the median from scratch every round, making it suitable for oblivious robots with no memory.
• Half‑Step Movement Guarantees Contraction – The simple geometric contraction argument works in one dimension and extends naturally to higher dimensions with appropriate norm‑based contractions (a direction for future work).

The authors complement the theoretical analysis with extensive simulations. They test various Byzantine strategies (e.g., constantly reporting extreme positions, random wandering, coordinated attacks) and different activation patterns (adversarial, random, round‑robin). In all cases, the correct robots converge within the predicted number of rounds, and the convergence speed matches the logarithmic bound derived analytically.

Importantly, the algorithm matches the known lower bound for the ATOM model, demonstrating that the same fault‑tolerance limit (3f + 1) is achievable even in the more hostile asynchronous CORDA setting. This result closes a gap in the literature and establishes that optimal Byzantine‑resilient convergence is possible without synchrony, memory, or randomization.

The paper concludes with several avenues for future research: extending the approach to multi‑dimensional spaces, relaxing the half‑step restriction to improve convergence speed, handling dynamic robot populations (join/leave), and implementing the algorithm on physical robot platforms to validate the bounded‑scheduling assumption in real‑world networks. Overall, the work provides a solid theoretical foundation and a practical algorithmic blueprint for building robust, fault‑tolerant swarm systems that can operate reliably even when a subset of agents behaves arbitrarily.