Stuck in Traffic (SiT) Attacks: A Framework for Identifying Stealthy Attacks that Cause Traffic Congestion

Recent advances in wireless technologies have enabled many new applications in Intelligent Transportation Systems (ITS) such as collision avoidance, cooperative driving, congestion avoidance, and traffic optimization. Due to the vulnerable nature of wireless communication against interference and intentional jamming, ITS face new challenges to ensure the reliability and the safety of the overall system. In this paper, we expose a class of stealthy attacks – Stuck in Traffic (SiT) attacks – that aim to cause congestion by exploiting how drivers make decisions based on smart traffic signs. An attacker mounting a SiT attack solves a Markov Decision Process problem to find optimal/suboptimal attack policies in which he/she interferes with a well-chosen subset of signals that are based on the state of the system. We apply Approximate Policy Iteration (API) algorithms to derive potent attack policies. We evaluate their performance on a number of systems and compare them to other attack policies including random, myopic and DoS attack policies. The generated policies, albeit suboptimal, are shown to significantly outperform other attack policies as they maximize the expected cumulative reward from the standpoint of the attacker.

💡 Research Summary

The paper introduces a novel class of stealthy attacks on Intelligent Transportation Systems (ITS) called “Stuck in Traffic” (SiT) attacks. In modern ITS, vehicles receive real‑time traffic information from smart traffic signs (digital signs, V2I beacons, etc.) and use this data to make routing decisions. By selectively jamming or corrupting a subset of these wireless signals, an adversary can manipulate the information presented to drivers, causing them to choose suboptimal routes that lead to congestion in targeted road segments.

Threat model and system assumptions
The attacker is assumed to have the capability to interfere with the wireless channel (e.g., via a low‑power jammer or a compromised roadside unit) and to possess at least partial knowledge of the system state—traffic densities, signal timings, and the content of the smart signs. The attack is “stealthy” because it does not require a blanket denial‑of‑service; only a few carefully chosen signals are altered, making detection by conventional IDS difficult.

MDP formulation
The authors model the attacker’s decision problem as a Markov Decision Process (MDP).

• State (s) – a vector describing the current traffic condition: vehicle counts on each road segment, the phase of traffic lights, and the information currently displayed on the smart signs.
• Action (a) – the selection of which sign(s) to jam or corrupt and at what power level. Because the attacker’s resources (energy, spectrum) are limited, the action space is a constrained subset of all possible signals.
• Transition (P(s′|s,a)) – the probabilistic evolution of traffic after the attacker’s interference. The authors estimate these transitions using microscopic traffic simulators (e.g., SUMO) combined with analytical flow models.
• Reward (R(s,a)) – a scalar representing the attacker’s objective: the increase in total travel time, the length of queues, or any cost metric that the traffic authority would seek to minimize. The attacker seeks to maximize the expected cumulative reward over an infinite horizon.

Policy synthesis via Approximate Policy Iteration (API)
Solving the MDP exactly is infeasible due to the high dimensionality of the state and action spaces. The paper therefore adopts Approximate Policy Iteration, a reinforcement‑learning technique that alternates between:

1. Policy evaluation – using sampled trajectories from the traffic simulator to estimate the value function (V^\pi(s)) for the current policy (\pi). Temporal‑difference (TD(λ)) or Monte‑Carlo estimators are employed.
2. Policy improvement – updating the policy parameters (linear weights or a shallow neural network) by a policy‑gradient step that maximizes the estimated advantage.

Through repeated iterations, the algorithm converges to a “sub‑optimal” yet highly effective attack policy that selects the most impactful signals to disrupt while keeping the number of interventions low.

Experimental evaluation
Four experimental scenarios are examined: (i) a single intersection, (ii) a multi‑intersection urban network, (iii) high‑density vs. low‑density traffic regimes, and (iv) a setting with adaptive signal control. The API‑derived SiT policy is compared against three baselines: random signal jamming, a myopic (greedy) policy that maximizes immediate reward, and a full denial‑of‑service (DoS) that blocks all communications.

Key findings include:

• The API‑based SiT policy consistently yields the largest increase in average travel time (15–30 % over baseline) while using fewer jamming actions than the DoS baseline.
• In multi‑intersection networks, the policy learns to “seed” congestion in a few critical links, causing a cascade of rerouting decisions that amplify overall delay.
• The myopic policy performs well only when the traffic state is static; it fails to anticipate the delayed effects of early interventions.
• Random jamming has negligible impact, confirming that indiscriminate interference is ineffective against a system that can adapt to occasional packet loss.

The authors also analyze detection difficulty. Because the SiT attacks involve low‑power, intermittent interference, standard spectrum‑monitoring tools and IDS that look for sustained high‑energy jamming often miss them.

Defense considerations
To mitigate SiT attacks, the paper suggests:

• Cryptographic authentication of V2I messages (e.g., PHY‑layer signatures) to prevent unauthorized modification.
• Cross‑validation of traffic information using heterogeneous sensors (cameras, radar, V2V exchanges) to detect inconsistencies.
• Anomaly detection on traffic flow patterns, flagging sudden, localized spikes in travel time that cannot be explained by normal demand fluctuations.

However, the authors acknowledge that these measures increase system complexity and cost, and that a fully robust solution may require a combination of secure communication, redundant sensing, and adaptive control algorithms.

Contributions and future work
The paper’s primary contributions are: (1) defining a realistic, stealthy attack vector that exploits the informational dependence of modern ITS, (2) formulating the attacker’s problem as an MDP and solving it with Approximate Policy Iteration, and (3) providing extensive simulation evidence that the derived policies outperform naïve attacks. Future research directions include real‑world test‑bed validation, game‑theoretic analysis of attacker‑defender interactions, and privacy‑preserving mechanisms for the data used in policy learning (e.g., differential privacy, blockchain‑based audit trails).

Overall, the work highlights a previously under‑explored security risk: the manipulation of traffic‑information services can be as disruptive as physical road blockages, and defending against such attacks will be essential as ITS deployments become ubiquitous.