Toward the Right Analytical Model and System Software for Autonomous Driving Systems: Open Problems and Research Directions

Jul 16, 2026

Upload by Hazel

Autonomous driving (AD) systems transform multi-rate, asynchronous sensor streams into vehicle actuation through callback graphs and middleware. Temporal correctness depends on end-to-end cause-effect chains: parallel localization and perception, timestamped data fusion, convergence at planning, and propagation through control to actuation. Processing variability on multicore CPUs and GPU-accelerated platforms makes rare deadline misses unavoidable, so safety relies on fail-safes such as minimal-risk maneuvers. The work surveys real-time analytical models and AD system software, identifies gaps across constraint units, timing metrics, resource models, execution-time variability, and safety integration, and outlines research directions to align theory with practical enforceable infrastructure.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages8

Reading Time20 min

Introduction

Motivation: timing correctness across sensor-to-actuation graphs

Limits of classical real-time assumptions in AD software

Toward the Right Analytical Model and System Software for Autonomous Driving Systems: Open Problems and Research Directions Atsushi Yano∗† and Takuya Azumi‡† ∗ Graduate School of Science and Engineering, Saitama University, Japan †TIER IV Incorporated, Japan ‡Academic Association (Graduate School of Science and Engineering), Saitama University, Japan arXiv :2607 .04 129v 1 [ cs . SE] 5 Jul 2026 Abstract—Autonomous driving (AD) systems continuously transform multi-rate and asynchronous sensor streams into vehicle actuation through graphs of callbacks, nodes, and middleware components. In such systems, temporal correctness cannot be characterized by the execution time or deadline of an individual task alone: localization and perception chains run in parallel, fuse data with different timestamps, converge at planning, and propagate through control to actuation. Moreover, the demand for high processing capability places AD systems on highperformance processors with multicore parallelism and GPU acceleration, where execution times vary strongly with the input scene, hardware state, and co-running work. Rare deadline misses at runtime therefore cannot be ruled out, and safety is preserved through fail-safe mechanisms such as the minimal-risk maneuver (MRM). This raises a two-sided question: what analytical models are needed to reason about timing in AD systems, and what system software is needed to realize, observe, and enforce those models on real platforms? On the analytical side, real-time research has evolved from periodic/sporadic tasks, directed acyclic graphs (DAGs), pipelines, mixed-criticality systems, and timer-/event-driven models toward end-to-end latency along cause-effect chains, data freshness, timing disparity, probabilistic timing, highest-criticality fail-safe operation, and early deadline-miss detection. On the system-software side, AD stacks and middleware, such as Autoware and ROS 2, expose both the opportunities and limitations of implementing analyzable timing behavior through executors, communication layers, tracing tools, and evaluation frameworks. This paper surveys these two lines of work and identifies the remaining gaps along five dimensions: units of timing constraints, timing metrics, resource models, executiontime variability, and safety integration. Rather than proposing a single new model or runtime, we formulate open problems and research directions for converging theory and practice: analytical models must move closer to AD reality, while AD system software must be reshaped into analyzable, enforceable, and safety-aware infrastructure. Index Terms—real-time systems, autonomous driving, analytical timing model, system software, ROS 2, Autoware, cause-effect chains, end-to-end latency, data freshness, probabilistic timing, deadline-miss detection, open problems. I. Introduction An autonomous driving (AD) system is a cyber-physical system that continuously transforms multiple sensor streams into vehicle actuation through a graph of callbacks rather than a single linear chain (Fig. 1) . This figure shows one representative Autoware configuration, a concrete instance rather than a fixed architecture. In it, each gray box is a Robot Operating System (ROS) 2 callback (or the node itself when the node holds a single callback), a dashed enclosure groups the callbacks of a multi-callback node, and solid edges carry publish/subscribe communication over a topic. Orange dotted edges instead carry data through a node-internal queue or the ROS 2 take API [1]; such an edge does not trigger the destination callback directly but affects its execution time and output. Each callback is triggered by a timer at a fixed period (marked with a timer icon), by a subscription to a single topic, or by the synchronization of several topics (a sync callback, using the ApproximateTime or ExactTime policy [2]) . Multi-rate sensors, such as multiple LiDARs, a global navigation satellite system (GNSS), and an ine

FAQs

Why can’t timing correctness in autonomous driving be judged by a single task’s execution time or deadline?

Because localization and perception run in parallel, fuse data with different timestamps, converge at planning, and then propagate through control to actuation. Timing correctness is a property of the entire sensor-to-actuation graph rather than one task.

What safety mechanism helps preserve correctness when deadline misses are rare but possible?

Safety is preserved through fail-safe mechanisms such as the minimal-risk maneuver (MRM), which provides an emergency stop behavior based on dead reckoning.

Which gaps does the paper identify for bridging analytical models and AD system software?

Remaining gaps are organized into five dimensions: units of timing constraints, timing metrics, resource models, execution-time variability, and safety integration. The paper frames open problems and research directions to make models closer to AD reality and system software analyzable and enforceable.

Jul 16, 2026

Upload by Hazel