Battery-Side Energy Efficiency of Wirelessly Connected Robots Using the Joint Offloading, Compression and Scaling Offloading and Scheduling Policy

Abstract

Battery capacity is the binding constraint on how long a fleet of mobile robots can work between charges. In wirelessly connected fleets, the robot battery is shared between three competing consumers: locomotion, onboard computation, and the radio. All energy figures in this paper refer exclusively to the robot battery; the edge server is mains-powered and outside the optimisation scope. Offloading perception to the edge server removes computation from the battery, but it consumes airtime on a shared wireless medium and a late result forces the robot to slow down, lengthening the mission and increasing locomotion energy drawn from the same cell. We develop a battery-side energy model that prices a missed perception deadline in joules through its effect on mission duration, derive the penalty in closed form as λ = 5.18 J, and propose joint offloading, compression and scaling (JOCS), a decentralised per-epoch policy that jointly selects the offloading decision, payload compression level, transmit power, and processor frequency for each pipeline stage. JOCS admits offload requests in decreasing order of battery energy saved per second of airtime consumed, subject to a feasibility test on the self-induced congestion. Simulated over 30 independent channel realisations, JOCS reduced robot battery energy by 22.6% relative to run-to-deadline local execution at eight robots per access point (95% CI: ±0.04%), with zero missed deadlines. At 24 robots, JOCS saved 19.5% against all-local and 31.9% against the best competing baseline (95% CI: ±0.7%). An extended ablation study with six component-removal variants isolates the contributions of compression, admission control, the latency-energy coupling term, and the robustness margin. A congestion-model sensitivity study with three alternative delay functions shows that the nominal saving is stable, while the congested-fleet advantage requires recalibration to the measured channel before deployment.