Opportunity for Visiting Researchers in Binary Neural Networks for Efficient Onboard AI for Earth Observation

Deadline: February 28, 2026 at 23:59 CET

The ESA Φ-lab CIN is now looking for visionary researchers and innovators in Transformative Technologies to conduct research in collaboration with the ESA Φ-lab and the ESA AI and Data science section in the field of Binary Neural Networks for Efficient Onboard Artificial Intelligence for Earth Observation. Find out more and apply now.

Main Information

🖊️ Topic: Binary Neural Networks for Efficient Onboard Artificial Intelligence for Earth Observation

🎓 For: PhD students or postdoctoral researchers (preferred) in Embedded Software, Computer Science, Artificial Intelligence, or related.

📍 Location: ESA Φ-lab, ESRIN, Frascati, Italy.

📅 Collaboration scheme and timeline: We look for a hybrid remote/onsite collaboration scheme with a foreseen timeline of 9 months – 1 year (ca). The Visiting researcher will have the possibility to be hosted at ESA ESRIN for maximum 6 months starting from June 2026.

🚀 Preferred collaboration kick-off date (remote): April 2026

Introduction and Context

This call is in the context of the European Space Agency (ESA) Φ-lab Collaborative Innovation Network (CIN), an initiative to foster collaborations among Earth Observation and Transformative Technologies experts. The CIN was created with the aim of attracting high-level researchers, distinguished advisors and key leaders in the above-mentioned fields to the ESA Φ-lab at the ESA Centre for Earth Observation (ESRIN), Frascati, Italy.

ESA Φ-lab is the division of ESA with the mission to accelerate the future of EO by means of transformational innovation, i.e., innovations with the potential to create or transform industries via new technologies, thereby strengthening the global competitiveness of the European EO industrial and research sectors.

Pi School, which operates in the field of AI and research innovation and has numerous academic contacts through its staff, is responsible for supporting ESA Φ-lab in this mission by creating, managing, coordinating, animating the ESA Φ-lab CIN.

Purpose and Scope

This specific call aims to engage with researchers, such as PhD candidates, postdoctoral researchers (preferred), or equivalent in Computer Science, Artificial Intelligence, Embedded Software, Electronics, or related that are eager for the investigation and validation of Binary Neural Networks (BNNs) for efficient deployment of edge AI models on board EO satellites.

Unlike traditional Convolutional Neural Networks (CNNs), which rely on high-precision floating-point operations, BNNs constrain both weights and activations to binary values (typically ±1). This extreme quantization enables bitwise operations (e.g., XNOR and bitcount) instead of costly matrix multiplications, resulting in orders-of-magnitude reductions in computational complexity and power consumption. Such aspect is promising for onboard satellite applications, especially for micro and nano-satellites.

Moreover, beyond efficiency, BNNs introduce a paradigm shift in AI verifiability. Their discrete and deterministic nature makes them amenable to formal methods, such as SAT/SMT solvers and symbolic execution, which are traditionally infeasible for full-precision deep networks. This enables the generation of certified guarantees about model behaviour — e.g., robustness to input perturbations, absence of unsafe outputs, or compliance with mission-critical constraints. Such guarantees are essential for trustworthy AI in safety-critical space systems, where explainability and certification are non-negotiable.

However, these potential advantages come at the price of increased complexity of training algorithms due to the extreme quantization that might significantly impact the performance of the models, require significant exploration and tune of training algorithms.

Because of that, this call is looking for research projects aiming to address the following research objectives:

• OBJ 1 - Performance Evaluation of BNNs for EO Tasks: Assess the accuracy and robustness of BNNs on representative EO tasks (e.g., image classification, segmentation, anomaly detection). Benchmark BNNs against traditional CNNs in terms of inference quality, latency, and resource usage (main objective)
• OBJ 2 - Power and Resource Efficiency Analysis: Quantify the computational and energy savings achieved by BNNs on representative embedded platforms. Demonstrate that BNNs can operate effectively without specialized hardware accelerators, making them suitable for a wide range of on-board systems.

Other secondary research objectives:

• OBJ 3 - Formal Verification and Certification Potential: Investigate the applicability of formal methods (e.g., SAT/SMT solvers, symbolic reasoning) to BNNs. Demonstrate the generation of formal guarantees (e.g., robustness, safety bounds) for selected EO tasks, supporting certifiable AI.
• OBJ 4 - Integration into On-Board Processing Pipelines: Explore the integration of BNNs into existing or emerging on-board processing architectures. Identify constraints, trade-offs, and design patterns for deploying BNNs in operational EO missions.
• OBJ 5 - Possible integration into the PyNAS framework: PyNAS is a Neural Architectural Search (NAS) tool aiming at optimizing the inference of AI models for in-orbit deployment. This is achieved by optimizing the AI model’s architecture according to specific optimisation goals by leveraging a hardware-in-the-loop approach. Currently, PyNAS does not support quantization or the deployment of BNNs, which are aimed to be integrated.

Collaboration partners

The research projects will be conducted in collaboration with the ESA Φ-lab and the ESA Data Science and AI section. The AI and Data Science Section implements the AI Competence Centre and serves as the Agency’s centre of excellence for advanced technologies, driving the development and exploitation of future space missions. It supports the core mission of the Technology, Engineering and Quality directorate by enabling European projects through excellent engineering and the provision of innovative technologies, processes, methods, tools, and applications. The Section also conducts targeted research in Artificial Intelligence, Data Science, Quantum Computing, and related fields.

Who are we looking for?

A highly motivated researcher in a related field to the topic of this call and a strong interest in (or preferred: experience in) Earth Observation. The researcher would join onsite at the ESA Φ-lab and conduct research in collaboration with other researchers from the ESA Φ-lab and ESA AI Data Science section for a period of 3 to 6 months (preferred). We will prefer candidates that are interested in continuing the collaboration after the visiting period ends, for example in the context of their PhD.

Key Activities:

• Identify SOTA techniques for the training of BNNs suitable for onboard AI EO use cases
• Gather and preprocess relevant EO data related to the identify use cases
• Implement and run models and training of BNN AI models
• Benchmark BNN models to CNN for the identified onboard AI use cases (nice to have: on relevant edge computing hardware for space applications)
• Contribute to joint publications with leading academic and industrial partners.

Requirements:

• Ongoing or completed PhD in Computer Science, Artificial Intelligence, Embedded Software, Electronics, or related
• Experience with Edge AI, AI models quantisation.
• Experience with Python and machine learning frameworks (e.g., PyTorch, HuggingFace, Weights & Biases).

Nice to have:

• Knowledge about EO and onboard AI, AI model optimisation, embedded SW, and embedded hardware.
• Interest in Earth Observation, Earth system science and scientific research.

Opportunities:

• Collaborate with top-tier organizations and leading researchers in AI, EO and Earth system science.
• Contribute to high-impact publications and innovative projects.
• Work in a dynamic, interdisciplinary research environment.

Type of Engagement

As a Visiting Researcher, you will be engaged through the Φ-lab Visiting Researcher Scheme: this is a collaboration performed at ESA Φ-lab with the goal of jointly advancing a specific research proposal of shared interest by leveraging shared expertise, resources and time within an ESA Φ-lab team. The collaboration is unpaid and ideal for researchers already affiliated with an academic institution or employed by a private company. Under specific conditions, ESA may provide coverage for travel and accommodation.

If selected, ESA will send you an invitation letter (see template here).

How to apply?

Interested candidates are invited to submit their application through the form below, including a detailed CV and any supporting documents or references. In addition, you will be asked to provide:

• a brief research proposal (max 5 pages) addressing the primary research objectives (e.g., OBJ1-2) and, potentially, one or more secondary research objectives (e.g., OBJ-3-5). The research proposal shall:
• State clearly the addressed primary and secondary research objectives and concisely propose a research methodology to address such objectives
• Propose the timeline for the project start, indicating preferred tentative on-site period
• If any, highlight relevant publications to the topic