Project Description

Abstract

The Bád Project, conducted under an ESA co-funded program starting in Q2 ’25, utilized Ubotica’s CogniSAT-6 edge AI platform to validate real-time maritime monitoring use cases. As part of the AI4EO (FutureEO Block 4) initiative, the program addressed critical needs for autonomy, data optimization, and low-latency decision-making in Earth Observation (EO). By successfully demonstrating on-orbit AI processing and change detection, the project validated a robust, trustworthy model for delivering actionable intelligence—such as identifying non-AIS "dark vessels" and tracking ship-to-ship transfers—within minutes, significantly reducing data latency compared to traditional ground-processed methods. This validation, supported by extensive engagement with public maritime stakeholders, provides a strategic foundation for evolving into a multimodal, autonomous satellite constellation capable of delivering resilient, global, live Earth intelligence.

Detailed description

Under an ESA co-funded program, in Q2 ’25, Ubotica commenced a project to explore the suitability of its on-orbit edgeAI demonstrator platform (“CogniSAT-6”) for validating and demonstrating real-time monitoring use cases in the maritime domain.

Part of ESA's AI4EO (now in FutureEO Block 4), this program addresses the need for real-time decision-making, autonomy, and data optimization in Earth Observation (EO). The growing interest in onboard intelligence, particularly edge AI, enables smart, software-defined satellites with greater autonomy. SmallSat missions using optical payloads, such as Φ-sat, have already demonstrated the operation of Machine Learning (ML) algorithms on EO platforms with off-the-shelf, ultra-low-power, AI-optimized chips (such as those used on CogniSAT-XE2).

Project Objectives and Background

The core objectives of the Bád Project were to:

1. Gain a comprehensive understanding of both commercial and public maritime domain stakeholder needs and opportunities for maritime monitoring using Space assets.
2. Understand the strategic priorities of these stakeholders regarding Space EO solutions.
3. Determine the critical next steps for commercializing EO data in the maritime domain.
4. Serve as a stepping stone toward the vision of autonomous intelligent satellites in a collaborative sensorweb.

The CogniSAT-6/HAMMER mission previously successfully demonstrated the technical feasibility of real-time, on-orbit AI processing, leveraging Ubotica's onboard intelligence expertise. Equipped with a 5-meter GSD hyperspectral imager, the CogniSAT-XE2 edge AI hardware accelerator, and a low-latency Inter-Satellite Link (ISL), the satellite analyzed captured raw imagery instantaneously. This allowed for the detection and extraction of high-value insights, delivering intelligence to ground-based servers within minutes.

Motivation

Maritime security and law enforcement face a complex challenge due to widespread missing, incorrect, or deliberately misleading information spread by vessels. Despite most traffic being compliant, achieving maritime domain awareness is difficult due to the sheer volume of work required to cover illicit activities and monitor sanctioned, uninsured, polluting, or threatening vessels.

As satellites are not constantly in contact with ground stations, images are usually captured and downlinked at the earliest available ground station contact. This delays data availability by hours or even days. The unique hardware capabilities of CogniSAT-6 allow the satellite to apply on-board intelligence and condense the full image into very small packets of actionable information. The information is small enough to be sent over a persistent low-bandwidth link, delivering information directly to any interested party. This can reduce the time of illegal activity alerts from multiple hours to minutes.

As an extension of this new opportunity, the Bád project focuses on leveraging small, on-satellite packets of information to enable a multitude of applications, with a particular emphasis on the actionability of the generated insights for end users and on-orbit change detection to enhance the autonomy of space assets.

Methodology

Identification and validation of stakeholder needs were performed through engagement with four diverse stakeholders in the maritime domain, who provided direct feedback on the actionable insights. The project aimed to bring together AI, EO data, and domain experts to implement and assess a robust, trustworthy onboard edge AI intelligence model for enhanced maritime situational awareness.

The key stakeholders have the following stakeholder focuses:

• Detecting and identifying vessels near critical infrastructure
• Detecting, identifying, and tracking vessels involved in narcotics smuggling, including suspicious movements like ship-to-ship transfers or drops at sea
• Gathering evidence for prosecuting vessels engaged in IUU fishing
• Monitoring of sanctioned vessels, or vessels with a potential threat to the environment

As part of the project's in-orbit demonstration, Ubotica completed two separate pilots, both demonstrating novel operational paradigms enabled by on-orbit intelligence.

The first pilot concerned low-latency insights delivery, while the second pilot concerned on-orbit change detection.

• Pilot 1: After initial stakeholder talks and determining, amongst other valuable information, their Areas of Interest (AoI). The CogniSAT-6 satellite is tasked to observe said AoIs. Following image acquisition, onboard intelligence processes the full image and reduces it to a set of small feature descriptors. This data size allows it to be sent over the ISL. The data is uniquely fused with on-ground available data, and the data is packed and shared in an agreed-upon, actionable format.
• Pilot 2: After tasking and acquiring 2 images over consecutive orbits over the same area of interest, on-board intelligence is run on both images. Features are extracted from both images immediately after imaging, and the features are matched to detect changes. Changes are downlinked to the ground through a persistent link.

The initial pilot successfully expanded Ubotica’s proven on-orbit capabilities by designing, implementing, and validating a comprehensive ground data flow. This flow ensures that downlinked data packages are thoroughly checked for errors, fused with available ground data, and then compiled into an agreed-upon format, in this case KML. This full flow, from image tasking to delivery to stakeholders, was validated using multiple in-orbit demonstrations.

The latter pilot is designed to increase efficiency by focusing only on relevant changes. For example, land features are known and of no interest. Instead, pattern matching is used on-orbit to detect only differences between the features of interest (here, vessels). The details of these changed features are then downlinked via ISL. This full flow was verified through multiple in-orbit demonstrations.

This capability offers significant advantages across sectors:

• Maritime Security: Extracted changes can autonomously trigger a third acquisition without ground intervention. This enables continuous tracking of IUU (Illegal, Unreported, and Unregulated) fishing fleets or vessels near critical infrastructure by dynamically adjusting internal schedules.
• Energy Sector: It instantly detects and transmits oil tanker movements from anchorages to offloading sites via the ISL. It also provides automatic monitoring of critical areas such as anchorages, chokepoints, and ports.

On-orbit change detection establishes a foundation for autonomous satellite constellations. By rapidly detecting a high-interest change, it facilitates "tipping and cueing" via ISL communication. The algorithm is sensor-agnostic, allowing its application across diverse use cases and heterogeneous constellations, ranging from ship-to-ship transfer monitoring to GPS-denied orbit determination.

Pilot 1 result

Image capture-to-final-analysis latency was just 1 hour, with a minimum cycle time of 11 minutes. The analysis successfully identified and confirmed multiple "dark vessels" (non-AIS-transmitting maritime targets). This proves its ability to detect non-cooperative targets.

Two example images are provided below, illustrating their use across different scenarios. The first image supports regional monitoring for use cases such as narcotics smuggling, critical infrastructure monitoring, fisheries transshipment monitoring, or the potential transshipment of oil by sanctioned vessels. The second image highlights a distinct use case for monitoring highly crowded maritime areas where vessels may be acting illicitly.

Pilot 1 Example - Critical infrastructure

Image 1 shows a monitored area for critical infrastructure, with several ships captured. These ships are not close to the underground infrastructure. One can see the land features and the positions of vessels; however, only the vessels' features are of actual use to the stakeholders, as they already know all their land features.

This is a crop of the full image, where critical infrastructure is imaged. One can see that no nefarious activity is being conducted in restricted areas. On the ground, only the coordinates of the restricted area and of any vessels need to be known.

Image 1: Image captured over critical infrastructure showing the detected vessels. Only the feature data of extracted vessels are downlinked in real time.

Pilot 1 Example - Crowded maritime area

Image 2 is an example of a monitored area for sanctioned vessels and illegal activities.

This image crop shows many vessels. To confirm their visibility to authorities, detections are cross-referenced with AIS data. Mismatches warrant further investigation. Comparing vessel features allows low-latency onboard processing. For the comparison, only the vessels' features are needed; thus, this investigation can be performed at much lower latency through onboard processing.

Image 2: Busy shipping area monitored for sanctioned and illegal vessels, showing the detected vessels. Only the feature data of extracted vessels are downlinked in real time.

Pilot 2 result

On-orbit matching was successfully achieved for 3 distinct pairs of images. Across different geographic extents, the 3 images show a wide range of applicability for on-orbit matching.

The diverse range of deployment areas enabled verification across a broad spectrum of the operational envelope. These areas, along with their general characteristics, were used:

• Gibraltar: A busy shipping lane with moving vessels and anchorages in the North and South.
• Dakar: Relatively calm waters containing only a handful of vessels.
• Port Said: Very busy anchorage area containing cloudy patches.

Pilot 2 example - On-orbit matching

Image 3 shows the matching run over Dakar. White boxes are plotted over ships detected in both images; the other vessels shown in the image are not detected in either image.

Broadcast AIS points are indicated in yellow for the matched vessels, showing that, in this example, a dark vessel was also tracked over multiple days.

Image 3: Matching results over Dakar. Image 1 (left), image 2 (right), with the matched vessels indicated in white. AIS signals are indicated in yellow.

Project Outcomes

Data shared and deliverables:

• Dataset Package:
• Hyperspectral imagery - 17 raw CogniSAT-6 images
• Extracted vessel insights
• Operational Datasets:
• Imagery and vessel detections shared with stakeholders captured between July and December 2025
• Stakeholder needs document
• In-orbit demonstration report

Publications

• “In-orbit demonstration of object-based change detection applied to the maritime domain”, C. Traba et al.