Closed-Loop Satellite Planning and Scheduling for Low-Latency Data Collection and Analysis With a Conversational Agent Interface

Closed-Loop Satellite Planning and Scheduling for Low-Latency Data Collection and Analysis With a Conversational Agent Interface

Salt Palace Convention Center, Salt Lake City, UT

We demonstrate a closed-loop geospatial intelligence (GEOINT) Tasking, Collection, Processing, Exploitation, and Dissemination (TCPED) pipeline that provides data and data analyses using commercial satellite imagery. The Cloudbased Automated Satellite Tactical TCPED (CASTT) software platform obtains data through automated interfaces with satellite operators, performs analyses of that data which drive automated re-tasking by Auria’s optimized collection planning software, and then issues those new follow-up tasks to the satellite operators. Automation shortens the timeline from the receipt of data that warrants follow-up data to its actual collection and delivery. CASTT leverages progress across the space industry to providing an integrated solution drawing from the best commercial offerings for satellite imagery and data analysis. We study currently realistic and theoretically achievable latency and present data and analysis from a real CASTT mission with closed-loop operation.

CASTT is driven by high-level user-specified missions, such as search-detect-track, rather than atomic tasks, such as data collection on individual Areas of Interest (AOIs). CASTT decomposes the high-level mission into initial tasks, e.g., to first perform data collection and analysis for search in relevant AOIs. It continuously and autonomously analyzes the data resulting from these tasks, e.g., to detect relevant vehicles in the AOIs and form initial tracks. CASTT then issues follow-up collection tasks to the satellites based on the analyses, e.g., to collect more data along the identified tracks. This results in lower latency and more relevant data. Meanwhile, it disseminates data and analyses directly to end users and enables the effective monitoring of larger areas through intelligent automation. In contrast, the current manually intensive satellite imagery intelligence process can be time consuming, laborious, and slow. Automation and optimization results in more timely information, better use of highly constrained resources, and more relevant, insightful intelligence.

Since CASTT is built with flexible and powerful APIs, it can be easily connected to new data providers, algorithm providers, or interfaces such as those leveraging emerging large language model (LLM) tools. We prototype a chatbot agent that allows users to interact with CASTT through natural language prompts.

CASTT tasking brokers data collection from different satellite operators. We first study the currently realistic latency from request-to-collection from a particular provider by studying responses to periodic queries through their API. We then study the theoretically achievable latency from data request to uplink to collection to downlink and delivery given real satellite tracks and a commercial ground station provider. We study the impact of brokering from a virtual constellation of the 7 satellite operators available through Auria. In a representative scenario, optimized brokering from the virtual constellation improved median (worst-case) latency by 39% (80%) relative to the next-best single satellite operator.

Finally, we demonstrate real CASTT automated data and analysis integration in a sample port monitoring mission. CASTT uses automated target recognition and pattern-of-life algorithms to detect ships in a port AOI and determines their typical behavior from historical imagery. When it detects a ship of interest is missing, it orders the appropriate follow-up tasking to collect data on another ports which the vessel frequents. In the follow-up image, it finds the ship of interest. We also demonstrate the configuration of this mission via chatbot, proving a pipeline from a natural language prompt to collected imagery, finished intelligence derived from AI/ML analyses, and cued follow-up imagery collection.