Enhancing Automotive User Experience with Dynamic Service Orchestration for Software Defined Vehicles

TL;DR

The paper addresses dynamic resource allocation for non-safety-critical BE applications in Software Defined Vehicles (SDVs) under changing onboard resources and V2X conditions by introducing Automotive eXperience Integrity Level (AXIL) and per-application runtime modes. It formulates a resource-aware, dependency-constrained optimization problem to maximize UX through the objective \\sum_i X[i, C[i]]\\ while respecting per-resource budgets \\sum_i P_{i, C[i], k} \\le R[k]\\ and mode dependencies, using a 3D resource matrix $P$ and a 2D AXIL score matrix $X$. The authors compare three solution approaches—a SAT-based baseline, a genetic algorithm, and a novel specialized heuristic that guarantees a valid solution in polynomial time—and demonstrate that the heuristic achieves near-optimal UX with real-time performance, making onboard runtime decisions feasible. The work offers a practical pathway toward scalable, UX-driven orchestration for thousands of apps in SDVs, balancing user experience with limited onboard resources and dynamic network conditions, and lays out clear directions for industrial validation and energy-aware extensions.

Abstract

With the increasing demand for dynamic behaviors in automotive use cases, Software Defined Vehicles (SDVs) have emerged as a promising solution by bringing dynamic onboard service management capabilities. While users may request a wide range of services during vehicle operation, background tasks such as cooperative Vehicle-to-Everything (V2X) services can activate on-the-fly in response to real-time road conditions. In this dynamic environment, the efficient allocation of onboard resources becomes a complex challenge, in order to meet mixed-criticality onboard Quality-of-Service (QoS) network requirements while ensuring an optimal user experience. Additionally, the ever-evolving real-time network connectivity and computational availability conditions further complicate the process. In this context, we present a dynamic resource-based onboard service orchestration algorithm that considers real-time in-vehicle and V2X network health, along with onboard resource constraints, to select degraded modes for onboard applications and maximize user experience. To enable dynamic orchestration, we introduce the concept of Automotive eXperience Integrity Level (AXIL) which expresses a runtime priority for non-safety-critical applications. This algorithm produces near-optimal solutions while significantly reducing execution time compared to straightforward methods as demonstrated by simulation results. With this approach, we aim to enable efficient onboard execution for a user experience-focused service orchestration.

Figures (5)

• Figure 1: Block diagram of the onboard orchestration platform. Our goal is to orchestrate the selection of runtime modes (including shutdown) for each application to remain within the available onboard, V2X, and edge offloading resources while maximizing the user experience (defined by our AXIL rating).
• Figure 2: Random problem instance generated with 20 applications and at most 5 modes. Each cell $(i,j)$ represents mode $j$ of $\mathcal{A}_i$ and is assigned a runtime priority $X_i^j$ (along with a vector of resource requirements $M_i^j$ defined in Section \ref{['sec:modes']}).
• Figure 3: Simulation results for (A) one particular parameter combination to study different algorithm options. $H$ is our specialized heuristic, $G_a^b$ is a genetic algorithm with $a$ the maximum number of generations and $b$ the number of generations to wait before stopping when the best score hasn't improved (if set), and $S$ is the exact solver based on Z3. We compare the (B) relative score, success rate, and (C) solve time of each algorithm for 1000 random instance repetitions. This figure serves as an illustration as the overall results greatly depend on the problem size. See Figures \ref{['fig:results_move_apps']} and \ref{['fig:results_mxp']} for a general performance assessment.
• Figure 4: Performance evaluation by moving the number of applications requested to launch from 5 to 100, with other parameters kept identical as Figure \ref{['fig:results_xp']}A. Each color follows the legend from Figure \ref{['fig:results_xp']}B. Lines represent the median value of 100 repetitions and corresponding areas the Q1 to Q3 quartiles.
• Figure 5: Multi-parameter performance comparison of our proposed heuristic algorithm $H$ with the (A) ILP solver $S$ and (B) genetic algorithm $G_{2k}^{100}$ chosen manually as the best competitor variant. Each cell represents the ratio of the median value from 30 instance repetitions on performance, solve time, or success rate. When a cell is empty, either $S$ or $G$ failed to find a solution for all repetitions while $H$ always finds all solutions. Hence, the success rate corresponds to $S$ (1h timeout) or $G$.