Distributed Algorithm for Collision Avoidance at Road Intersections in the Presence of Communication Failures

Vehicle-to-vehicle (V2V) communication is a crucial component of the future autonomous driving systems since it enables improved awareness of the surrounding environment, even without extensive processing of sensory information. However, V2V communication is prone to failures and delays, so a distributed fault-tolerant approach is required for safe and efficient transportation. In this paper, we focus on the intersection crossing (IC) problem with autonomous vehicles that cooperate via V2V communications, and propose a novel distributed IC algorithm that can handle an unknown number of communication failures. Our analysis shows that both safety and liveness requirements are satisfied in all realistic situations. We also found, based on a real data set, that the crossing delay is only slightly increased even in the presence of highly correlated failures.

💡 Research Summary

The paper addresses the critical problem of safely coordinating autonomous vehicles at road intersections when vehicle‑to‑vehicle (V2V) communication is unreliable. While many existing intersection‑crossing (IC) protocols assume bounded communication delays or rely on a centralized manager, this work proposes a fully distributed algorithm that tolerates an unknown number of consecutive packet‑loss events. The authors consider two autonomous cars approaching a four‑section intersection. Each vehicle is equipped with GPS/DGPS, an IMU, and a DSRC‑based V2V unit, and periodically broadcasts a heartbeat (HB) message containing its unique identifier, type, and state estimate (position, velocity, acceleration).

When a vehicle reaches the capture area—defined as the region where it can no longer stop before entering the intersection—it sends an ENTER message; after it has completely traversed the intersection it sends an EXIT message. The communication model assumes that a burst of receive‑omission failures may occur for any vehicle (denoted f ≥ 0), but that the vehicle will successfully receive its ENTER and EXIT messages in the (f + 1)‑th attempt. At least one HB is guaranteed to be received before the algorithm starts, which is realistic given the larger communication range compared to sensor range.

The core of the algorithm is the computation of a Mean Time to Intersection (MTI) for each vehicle:

 τₜ^{MTI} = (‑vₜ + √(vₜ² + 2aₜ(x_S ‑ x̂ₜ))) / aₜ

where vₜ, aₜ, and x̂ₜ are the current velocity, acceleration, and estimated longitudinal position, respectively, and x_S is the geometric centre of the intersection. The vehicle with the smaller MTI obtains the right‑of‑way; ties are broken by comparing unique identifiers. Importantly, MTI is calculated only from the mean estimate, not from its variance, because priority selection is not safety‑critical.

Once both vehicles have exchanged their ENTER messages (or after the guaranteed reception of the (f + 1)‑th message), each can determine the potential collision area (COL), which may be any subset of the four intersection sections or empty (e.g., both turning right). The algorithm then distinguishes two control regimes:

1. SAFE_CTRL – before agreement, each vehicle continues with its current acceleration a_{PR} (typically the most recent commanded value).

2. MAIN_CTRL – after agreement, if COL is empty or the MTI difference exceeds a safety threshold τ_TH (derived from vehicle speed and minimum safe distance), the vehicle proceeds with a_{PR}. Otherwise, it must decelerate to avoid collision. The required deceleration is computed as

 a_{NOPR} = aₜ ‑ 2 D / (τ_{t+1}^{C_{OL}})²

where D is at least the width of the collision area and τ_{t+1}^{C_{OL}} is the worst‑case remaining time to reach COL assuming constant acceleration.

Safety (no collisions) is proved by showing that two vehicles can never occupy the same subsection simultaneously because the MTI‑based ordering guarantees exclusive access, and the guaranteed reception of ENTER/EXIT messages prevents stale information from causing unsafe decisions. Liveness (finite crossing time) follows from the fact that every vehicle eventually receives the necessary messages and, after the capture area is passed, the algorithm permits forward motion without indefinite waiting.

The authors evaluate the protocol using a real‑world urban traffic dataset. They simulate varying packet‑loss probabilities and burst lengths up to ten consecutive losses, including highly correlated loss patterns. The results indicate that the average crossing delay increases by only about 5–7 % compared with an ideal loss‑free scenario, and the collision rate remains zero. This demonstrates that the MTI‑based priority mechanism is robust to severe communication degradation and that the distributed approach incurs negligible performance penalties relative to centralized traffic‑light control.

In conclusion, the paper delivers a practical, fault‑tolerant distributed intersection‑crossing algorithm that requires only minimal V2V messaging, no infrastructure support, and works under realistic, unpredictable communication failures. The authors suggest future extensions to multi‑lane, multi‑vehicle intersections, incorporation of malicious‑behavior detection, and on‑road experiments with actual autonomous platforms.