Equinox: Decentralized Scheduling for Hardware-Aware Orbital Intelligence

Researchers Ansel Kaplan Erol and Divya Mahajan have introduced Equinox, a novel decentralized scheduling system designed specifically for orbital AI platforms on Earth-observation satellites. The system addresses the critical challenge of managing time-sensitive tasks on resource-constrained satellites that must balance intermittent solar energy harvesting with computational demands. Unlike traditional schedulers that rely on static priorities, Equinox creates a dynamic 'marginal cost of execution' by compressing multiple constraints—including battery charge levels, thermal headroom, and task backlog—into a single, state-dependent value derived from barrier functions that escalate near safety limits.

This local cost signal serves as a coordination primitive across entire satellite constellations. When a task's perceived value exceeds the current execution cost, it proceeds; otherwise, it's shed. Crucially, if one satellite's local cost exceeds a neighbor's, tasks can be dynamically offloaded via inter-satellite links, enabling distributed load balancing without complex routing protocols or global state synchronization. The researchers validated Equinox through multi-day simulations of a 143-satellite constellation, using performance data grounded in actual NVIDIA Jetson Orin Nano hardware measurements.

The results demonstrate significant improvements over conventional approaches: Equinox boosted scientific data goodput by 20% and image-processing throughput by 31% while maintaining 2.2 times higher mean battery reserves. Under periods of intense computational demand, the system achieved 5.2 times the execution rate of static schedulers by gracefully shedding low-value work rather than collapsing under contention. This represents a major advancement for orbital edge computing, where efficient resource management directly translates to mission success and operational longevity.