Loft Orbital to test AI models on spacecraft for Earth observation

· Source: SpaceNews · Field: Technology & Digital — Artificial Intelligence & Machine Learning, Robotics & Autonomous Systems, Environmental Science & Earth Systems · Depth: Advanced, quick

Summary

Loft Orbital, in collaboration with NASA's Jet Propulsion Laboratory, is testing artificial intelligence on spacecraft to enhance Earth science monitoring. This initiative, part of the NASA-funded Federated Autonomous Measurement (FAME) project, began this month with JPL's AI software running on a Loft Orbital spacecraft. Further tests are scheduled for 2027 and 2028. The primary goal is to automate the "tip-and-cue" process, enabling real-time identification and detailed observation of Earth features directly on orbit, eliminating the need to downlink large data volumes. This approach leverages AI models trained on extensive data, capable of recognizing features without explicit instructions. Key challenges include integrating satellite infrastructure for real-time image processing and utilizing high-performance, multimodal AI models small enough for space hardware constraints. Loft Orbital plans to expand these capabilities with its Altair series of 10 satellites, featuring multiple sensors, edge computing, and intersatellite links for rapid insights.

Key takeaway

For AI Engineers developing space-based applications, this collaboration highlights the increasing viability of deploying advanced AI models directly on orbit. You should prioritize optimizing multimodal AI models for constrained edge hardware and explore intersatellite communication protocols to enable real-time "tip-and-cue" operations. This shift allows for immediate data analysis and action, significantly reducing latency for critical Earth observation tasks like wildfire detection or maritime surveillance, offering substantial commercial and governmental value.

Key insights

Onboard AI processing on satellites enables real-time Earth observation and rapid response without extensive data downlink.

Principles

Method

AI models, trained on large datasets, process satellite imagery in real-time to identify features of interest, then cue other spacecraft via intersatellite links for follow-up.

In practice

Topics

Best for: Computer Vision Engineer, AI Scientist, Research Scientist, AI Engineer, AI Hardware Engineer

Related on AIssential

Open in AIssential →

Editorial summary, takeaway, and curation by AIssential. Original article published by SpaceNews.