Injecting Hallucinations in Autonomous Vehicles: A Component-Agnostic Safety Evaluation Framework

TL;DR

This paper tackles the challenge of safety assurance for autonomous vehicle perception by reframing perception faults as hallucinations—observable, sensor-agnostic anomalies in MP outputs. It introduces a configurable hallucination injection (HI) framework embedded in an open-source simulator to stress-test AV safety across six hallucination types and three perception domains. Through 18,356 simulations in an unsignalized intersection scenario, the study demonstrates that hallucinations markedly increase collision risk (e.g., accident odds up to 8.53 at higher probabilities) and reduce inter-vehicle spacing, with certain types (missed detections and blind regions) and conditions (latency, angular drift, persistence) contributing most severely. The HI framework offers a scalable, interoperable tool for system-level AV safety evaluation, with potential to guide safety engineering, monitoring, and regulatory validation across diverse MP architectures and future AV platforms.

Abstract

Perception failures in autonomous vehicles (AV) remain a major safety concern because they are the basis for many accidents. To study how these failures affect safety, researchers typically inject artificial faults into hardware or software components and observe the outcomes. However, existing fault injection studies often target a single sensor or machine perception (MP) module, resulting in siloed frameworks that are difficult to generalize or integrate into unified simulation environments. This work addresses that limitation by reframing perception failures as hallucinations, false perceptions that distort an AV situational awareness and may trigger unsafe control actions. Since hallucinations describe only observable effects, this abstraction enables analysis independent of specific sensors or algorithms, focusing instead on how their faults manifest along the MP pipeline. Building on this concept, we propose a configurable, component-agnostic hallucination injection framework that induces six plausible hallucination types in an iterative open-source simulator. More than 18,350 simulations were executed in which hallucinations were injected while AVs crossed an unsignalized transverse street with traffic. The results statistically validate the framework and quantify the impact of each hallucination type on collisions and near misses. Certain hallucinations, such as perceptual latency and drift, significantly increase the risk of collision in the scenario tested, validating the proposed paradigm can stress the AV system safety. The framework offers a scalable, statistically validated, component agnostic, and fully interoperable toolset that simplifies and accelerates AV safety validations, even those with novel MP architectures and components. It can potentially reduce the time-to-market of AV and lay the foundation for future research on fault tolerance, and resilient AV design.

Figures (15)

• Figure 1: Hallucination produced by a 360° image reconstruction algorithm
• Figure 2: MP system model
• Figure 3: Taxonomy of dimensions and persistence of MP faults
• Figure 4: Illustration of observable consequences of CCD partial fault (left) and complete fault (right)
• Figure 5: Camera calibration drift