TII OrbitSight Challenge

TII has identified a specific problem within the Space Tech industry that requires innovative approaches beyond existing solutions.

Space Situational Awareness (SSA) involves detecting, tracking, and forecasting the movement of objects in Earth's orbit. It is crucial for protecting space assets and preventing collisions, yet challenging due to the vast number of objects, their high-velocity interactions, and difficult lighting conditions.

The main challenges are from the massive and increasing number of orbital debris, and difficult lighting conditions. Existing sensor systems are only able to track and catalog debris larger than about 10 cm in size. The collision of large debris could result in thousands of trackable pieces of debris and tens of thousands of smaller pieces of debris that are not trackable, especially under low-light, noisy conditions.

Traditional optical sensing technologies have proven their value over the years, but they have limitations in the conditions space demands: low-light environments, fast-moving objects, and the need for real-time response.

With recent advancements in vision sensor technologies, Neuromorphic Vision Sensor (NVS) have emerged as a promising solution for SSA. However, realizing their full potential requires the right algorithm that is designed and optimized to handle the unique nature of neuromorphic data.

Unlike conventional cameras that capture frames at fixed intervals, NVS sensors react to change - detecting movement at the microsecond level, with significantly lower power consumption and latency.

The potential is clear. But here is the challenge: Raw NVS data is asynchronous, event-driven, and noisy. Currently, there are no widely adopted, high-performance algorithms capable of processing that raw data into reliable, real-time detection and tracking of fast-moving objects.

To support this challenge, a real NVS dataset will be provided, captured by a sensor mounted on a high-end 0.8-meter diameter telescope at the Abu Dhabi Quantum Optical Ground Station (ADQOGS).

This challenge aims to develop AI and machine learning solutions capable of processing NVS data end-to-end: from raw sensor input all the way to real-time detection, tracking and visualization of space objects.

The best solutions will address these limitations while remaining practical and deployable in real-world environments, technically robust, scalable, and ready to make a real impact.

This is crucial for protecting satellites and preventing collisions, particularly as the satellite count rapidly grows.

TII is already investigating the use of AI with Neuromorphic Vision Sensors (NVS) to detect dynamic objects, as NVS possesses key technical advantages compared to traditional cameras.

Solutions may range in TRL but should be practical and clearly articulate how they can be used – Purely conceptual or theoretical solutions are not eligible.

The winning solution will be further developed by TII to enhance national capabilities in space debris detection and monitoring.

TII requests technical documentation and source code to be provided as part of the competition.

All submissions will be assessed based on the following criteria:

• Technical Innovation (AI Approach)
  
  • Definition: Innovation and methodological depth of the proposed AI model.
  • Description: Evaluates the novelty of the model architecture (e.g., SNNs, graph-based NNs, hybrid event-frame models), the rationale behind design choices, and the suitability of the approach for processing sparse event-based inputs. Strong submissions include an ablation analysis of alternative models to justify the optimal solution quantitively and qualitatively.

• Detection Accuracy (on test data)
  
  • Definition: Detection performance of the proposed AI model on unseen/new dataset.
  • Description: Evaluated using key detection metrics including mAP (mean Average Precision), precision, recall, and F1 score.
  
  
• Real-time Performance
  
  • Definition: Inference efficiency of the proposed AI model on the evaluation hardware.
  • Description: Evaluated using inference latency, throughput, and total runtime. Strong submissions achieve real-time performance (latency under 40 ms end-to-end).
  
  
• Documentation / Visualization & Reporting
  
  • Definition: Quality of the solution's documentation and visual reporting.
  • Description: Evaluated based on the proposal report, README, and supporting visualizations (e.g., architecture diagrams, training curves, detection examples, failure cases). Strong submissions clearly explain methodology, justify design choices, and present results in a reproducible and accessible manner.
  
  
• Team Competency / Solution Articulation
  
  • Definition: Expertise of the team and clarity in communicating the solution.
  • Description: Evaluated based on the relevance of team members' background to the problem, depth of technical understanding shown in the proposal and the pitch presentation, and ability to clearly articulate design choices, limitations, and results.