ARTFEED — Contemporary Art Intelligence

SAT-Edge-Agent: Hardware-in-the-Loop Edge-Agent for Onboard Satellite Intelligence

ai-technology · 2026-08-06

A recent preprint on arXiv (2608.03728) presents SAT-Edge-Agent, an edge-agent system that integrates hardware-in-the-loop (HIL) technology for onboard satellite intelligence. This system operates on a commercial ARM-based heterogeneous edge system-on-chip. It manages a browser workspace alongside a FastAPI agent, utilizing a local OpenAI-compatible language service and a YOLO-style oriented-object-detection endpoint that provides structured results backed by FAIR1M metadata. Two fixed FAIR1M workloads—one involving a single image and another with two images in series—were executed 20 times each, achieving a perfect completion rate. The average latency for the single-image task was 29.353 seconds, while it was 60.937 seconds for the two-image task, with P95 values of 31.166 and 66.882 seconds, respectively. The average detection times were 861.386 milliseconds and 1510.920 milliseconds, representing only 2.93% and 2.48% of the Full-Agent averages. This system is designed to convert mission intent into local tool calls, reveal execution states, and produce machine-readable artifacts while adhering to communication and power limitations. The full paper can be accessed on arXiv.

Key facts

  • SAT-Edge-Agent is a hardware-in-the-loop edge-agent system for onboard satellite intelligence.
  • Deployed on a commercial off-the-shelf ARM-based heterogeneous edge system-on-chip.
  • Coordinates a browser workspace and FastAPI agent with a local OpenAI-compatible language service.
  • Uses a YOLO-style oriented-object-detection endpoint returning FAIR1M metadata-backed structured results.
  • Two fixed FAIR1M workloads (single-image and serial two-image) were repeated 20 times each, with 20/20 completions.
  • Mean Full-Agent latency: 29.353 s (single-image) and 60.937 s (two-image); P95: 31.166 s and 66.882 s.
  • Mean detector time: 861.386 ms and 1510.920 ms, only 2.93% and 2.48% of Full-Agent means.
  • Designed to translate mission intent into local tool calls and return machine-consumable artifacts under constraints.
  • Paper available on arXiv with identifier 2608.03728.

Entities

Institutions

  • arXiv
  • FAIR1M

Sources