COMPASS: A multi-objective computational offloading framework for edge-cloud continuum in real-time industrial telemetry

This paper presents COMPASS (COmputationally-Managed Pareto-Aware Scheduling System), a multi-objective computational offloading framework for real-time telemetry in industrial edge-cloud continuum architectures. COMPASS jointly optimizes latency, energy consumption, and data fidelity through a lightweight Pareto-front approximation heuristic with O ( n log n ) per-decision complexity and empirically validated convergence. The framework integrates three components on a four-tier hierarchical architecture spanning on-vehicle edge processors, trackside fog nodes, regional servers, and cloud infrastructure: (C1) a multi-objective offloading algorithm with Pareto-aware utopia-point selection, (C2) an entropy-based adaptive sampling strategy with a formal fidelity bound, and (C3) a multi-tier graceful degradation protocol with bounded performance under node failures. The system is field-validated on a high-speed motorsport platform processing over 100 sensor channels at aggregate rates exceeding 50 Mbps under intermittent 5G connectivity. Field deployment demonstrates sub-50 ms 99th-percentile latency for critical edge-fog paths (22.4 ms mean), 99.7% data delivery reliability, and 99.92% system availability over 120 h across 2.4 TB of telemetry. Separately, simulation-based algorithm evaluation against seven baseline offloading strategies—including NSGA-II and a capacity-aware greedy—shows COMPASS achieves the best hypervolume indicator with 0% deadline misses for safety-critical tasks, while the field-validated infrastructure reduces on-vehicle energy by 73% relative to commercial baselines. The architecture transfers to smart manufacturing, autonomous vehicle sensor fusion, and industrial IoT deployments requiring predictable quality of service.