AI-Powered SSA: Anomaly Detection with Gemma and Explainable AI
The increasing congestion of orbital space necessitates advanced Space Situational Awareness (SSA) capabilities to identify anomalous object behavior, predict potential collisions, obtain insights into mission profiles, and assess potential threats. This challenge focuses on developing an intelligent system that uses synthetic or real-world space object telemetry to detect and categorize anomalous events. Participants will design, implement, and evaluate a machine learning pipeline capable of identifying unusual orbital maneuvers, conjunction events, or non-standard satellite operations. The system should go beyond simple detection by leveraging large language models to provide actionable, explainable insights into the nature and potential implications of detected anomalies, crucial for both civil and defense space tracking efforts. The solution will integrate a robust anomaly detection model, potentially built with or informed by Gemma 2 principles, processing diverse sensor data (e.g., TLEs, radar, optical observations). Furthermore, it will utilize the Langroid framework to build an interactive agent that queries the anomaly detection system and uses the Anthropic API to generate rich, context-aware explanations for identified anomalies. This will enable human operators to quickly understand complex situations and make informed decisions, directly supporting enhanced Space Situational Awareness and threat assessment capabilities.
What you are building
The core problem, expected build, and operating context for this challenge.
Develop an AI system using machine learning for anomaly detection, integrated with a Langroid agent and Anthropic API for explainable threat assessment in Space Situational Awareness.
How work is evaluated
The evaluation will assess the accuracy of anomaly detection, the effectiveness of the interactive agent, and the quality and relevance of the AI-generated explanations. Participants will provide their Python code and a demonstration of the interactive system.
Challenge datasets
Public datasets, plus the private datasets you have access to.
What you should walk away with
Develop a conversational AI agent using Langroid for interactive anomaly querying.
Integrate open-source large language models (or their principles, like Gemma 2) for data interpretation and feature engineering.
Utilize advanced LLM APIs (Anthropic API) for generating detailed, context-rich explanations of complex space events.
Master time-series anomaly detection techniques for orbital data.
Implement a robust machine learning pipeline for real-time SSA data analysis.
Reference links and supporting material
Sample data for 2 tasks
Challenge setup
The evaluation will assess the accuracy of anomaly detection, the effectiveness of the interactive agent, and the quality and relevance of the AI-generated explanations. Participants will provide their Python code and...
Challenge input
JSON array of time-series orbital telemetry data (e.g., {'timestamp': ..., 'sma': ..., 'inc': ..., 'ecc': ...}) with a 'true_anomaly_label' field.
Agent execution
No agent tool is listed for this challenge.
Evaluated output
JSON array of predictions for each data point (0 for normal, 1 for anomaly) and an associated confidence score.
- Ensures the submitted Python code runs without fatal errors and handles typical input cases.
- Verifies that the Langroid agent can process queries and return responses within a reasonable time limit (e...
- F1 Score AnomalyDetection target: 0.85
- Python execution harness
Summary shows the first task, up to three tools, and up to three evidence items. Open the technical recipe to view all configured items.
View technical recipe
Configured tools
No tool records are attached.
Evaluation contract
- Ensures the submitted Python code runs without fatal errors and handles typical input cases.
- Verifies that the Langroid agent can process queries and return responses within a reasonable time limit (e.g., <5 seconds).
Metrics
- F1_Score_AnomalyDetection · Target 0.85
- ExplanationRelevanceScore · Target 4
- ActionabilityScore · Target 3.5
Workflow tasks
- Task 1
Input: JSON array of time-series orbital telemetry data (e.g., {'timestamp': ..., 'sma': ..., 'inc': ..., 'ecc': ...}) with a 'true_anomaly_label' field.
Output: JSON array of predictions for each data point (0 for normal, 1 for anomaly) and an associated confidence score.
Execution harness: Python
- Task 2
Input: JSON object containing a detected anomaly's features and metadata (e.g., {'anomaly_id': 'AX123', 'features': {'sma_change': 50, 'inc_change': 0.1}, 'detection_time': '2025-11-28T10:00:00Z'}).
Output: JSON object with a 'explanation_text' field (string) and 'action_recommendations' (array of strings).
Execution harness: Python
Recipe freeze marker
No recipe freeze time is recorded.