About the project

The SHARCEX project focuses on improving underwater operations by integrating advanced autonomous underwater vehicles (AUVs) with human divers. The goal is to enhance safety and efficiency in extreme underwater environments such as defense, rescue operations, and law enforcement.

Key technologies being developed and integrated include:

The project progresses through multiple phases:

  1. Development of AI models
  2. System integration
  3. User interface design
  4. Operational testing

KTH leads the project in collaboration with FMV and Saab, aiming to demonstrate robust AUV-Diver collaborations validated through extensive simulations and experiments.

Background

Underwater environments are among the most challenging operational domains due to their unpredictability, extreme pressure conditions, limited visibility, and communication constraints. Human divers working in these environments face significant safety risks, particularly in defense, search-and-rescue, law enforcement, and infrastructure inspection scenarios. While Autonomous Underwater Vehicles (AUVs) have been deployed in these fields, their potential remains largely untapped due to limitations in real-time adaptability and human interaction.

Current AUV systems often operate independently, following pre-programmed missions with limited real-time decision-making capabilities. This restricts their usefulness in dynamic and high-risk situations, where divers must quickly adapt to changing conditions, assess threats, and make complex decisions. There is a clear need for AUVs that can function as intelligent, real-time assistants, enhancing human capabilities rather than merely executing pre-set tasks.

The SHARCEX project addresses this gap by developing a next-generation human-robot collaboration frameworkfor underwater operations. By integrating AI-driven AUVs with human divers, the project aims to create a synergistic system where both human and machine leverage each other’s strengths.

Crossdisciplinary collaboration
The project integrates expertise from multiple fields:

Collaboration partners:

Principal Investigators (PIs)

About the project

Objective
With the OrganoFeed project, we aim to leverage our joint expertise regarding microfluidic engineering & integration, and predictive algorithms development, to help address a core problem in biomedical research: reproducibility. Specifically, we aim to greatly reduce the variability of organoid cultures, which otherwise hold great promise for improving both fundamental research and drug development, by shifting the paradigm from a homogenous chemical environment to individualized, data-driven feedback control.

Background
Organoids are miniaturized, self-assembled, and self-organized cellular constructs. They can recapitulate key morphology, cellular composition, and biological function of human organs, improving greatly upon the simplistic mono-cellular models in use for early drug development. At the same time, organoids’ human origin avoids the species mismatch inherent to animal testing, which currently contributes significantly to poor translatability from drug candidates to human clinical trials (not to mention inherent ethical concerns). Last but not least, being derived from individual human donors’ cell samples, organoids can be used to model both fully personalized response as well as true population-level sampling. Organoids are, however, sensitive to even small variations in their culture conditions over the often weeks-long course of their maturation, resulting in high variability of morphology, cell composition, and function.

Current mitigation approaches have focused on providing more homogenous conditions. We propose instead an entirely different approach, based on feedback-driven control of the chemical environment at the level of each individual organoid. This ability to generate highly homogenous organoid populations should further increase organoids’ attractiveness in replacing both overly simplistic cell models as well as ethically and functionally suspect animal models with something more meaningful.

Crossdisciplinary collaboration
In this project, we are establishing a new cross-disciplinary and complementary collaboration:

About the project

Objective
RECOPS aims to provide evidence-based insights into the benefits of open-source software in the power sector. The project focuses on identifying impactful areas for open-source applications, such as HVDC control, distributed renewable energy (DER) simulation, and hardware operations. Through case studies and a novel assessment methodology, RECOPS will evaluate system robustness and cost benefits from open-source approaches. The project is structured into three work packages, incorporates stakeholder engagement, simulations, and interdisciplinary collaboration.

Background
Open-source software has gained widespread use in industries like robotics and general software engineering but remains underutilized in the power sector. While academic initiatives exist, their limited maintenance and lack of industry relevance hinder broader adoption. The power sector’s need for robustness and focus on proprietary, closed software solutions further constrain open-source integration. However, there could be benefits with open-source software to enhance cost-effectiveness and robustness in energy systems.

Crossdisciplinary collaboration
The researchers in the team represent the KTH School of Industrial Engineering and Management, Unit of Energy Systems and the KTH School of Electrical Engineering and Computer Science, Division of Electric Power and Energy Systems.

About the project

Objective
This project aims to develop a Risk Prediction Tool to use as an early warning system in primary care, aimed at cancer using Colorectal Cancer as the first test case. The tool will be based on data collected at primary care units, such as care visits such as history, symptoms, medication, lab tests and results, as well as free text from physician notes. The tool will use machine learning (ML), large language models and advanced statistical analyses to assess the risk of patients having an underlying yet undiagnosed CRC. Despite the promise and potential of primary care data to assess risk and improve screening, very few studies use this potential to better inform diagnostics.

Existing risk assessment tools designed for use in primary care have been developed and validated with a limited dataset (ICD-10 codes that were present in primary care datasets, covering a smaller percentage of patients), but are not yet in clinical use. The project aims to gain an understanding of the most salient data points for assessing risk of CRC in primary care, as well as insight into the potential benefits of such analysis for the healthcare system. Through this analysis, the project also aims to obtain a validated, novel risk prediction tool for CRC, which can be widely deployed and feed into national guidelines for colorectal cancer. By using CRC as the initial test case, the project aims to develop methods and tools through the results of the project which can be modified for other cancers and applications.

Background
A majority of patients with cancer present with early symptoms in primary care, before receiving their diagnosis. Despite the fact that Sweden has fast track for cancer diagnosis since 2015 (cancer care pathway, CCP) primary care physicians lack decision support to assess cancer risk for individual patients. Colorectal cancer (CRC) is the third most common cancer in Sweden. Despite screening, most patients with CRC are diagnosed after symptom presentation. The survival is excellent in patients diagnosed with CRC limited to the bowel wall (stage I-II) and intermediate when it has spread to regional lymph nodes (stage III) while the prognosis is poor in the approx. 25% who are diagnosed with CRC with distant metastases (stage IV).

Cross-disciplinary collaboration
The project is a highly inter-disciplinary collaboration between KTH, Regional Cancer Centrum Stockholm-Gotland, Regional Cancer Centrum Väst and Karolinksa Institutet. The project team comprises primary care clinicians, statisticians, modelers, software engineers and planners at RCC.

About the project

Objective
AsthmaTuner is an existing remote digital tool owned by MediTuner AB in Sweden. AsthmaTuner consists of a handheld spirometer connected to a mobile app that can be used by patients with asthma and other respiratory diseases to diagnose and manage their health condition.

Explainable Artificial Intelligence (XAI) has high potential to enhance existing digital tools previously developed without AI or with black-box AI. The existing AsthmaTuner system does not use any AI so far, and naturally the scope of using black-box AI has a high potential. In this proposed project, we pursue a significant step ahead. We will explore how explainable AI (XAI) will enhance both the diagnostic and management capabilities of the AsthmaTuner system. That means, XAI will help both patient users and their health care providers further improve asthma care, with the target of reduced need for unnecessary health care visits, or costly and potentially deadly asthma attacks or exacerbations. The XAI-based digital tools are expected to prevent further health disparities and will include the addition of social determinants of health to the design, implementation, and evaluation to AsthmaTuner to improve the performance.

Background
Asthma is a large-scale health care problem, affecting around 10% of the European population, and moreover; asthma and other respiratory diseases are increasing due to environmental deterioration worldwide. Asthma is a complex problem, which is characterized by airway inflammation and respiratory symptoms of wheezing, dyspnea, chest tightness and cough that vary over time and in intensity, together with variable expiratory airflow limitation. Potentially effective ways to address asthma care are to better understand the day-to-day lung function (a longitudinal data) and asthmatic symptoms of individuals and more effectively identify, diagnose, and manage asthma by exploring potential disparities in symptoms, lung function, and outcomes that may exist by social determinants for patients. AsthmaTuner (AT) is the existing, validated digital tool in this regard.

The general purpose of the project is to design an explainable AI (XAI)-based smart AT system that uses a mobile app and a handheld spirometer. We refer to the proposed project as “A3S: AI-based Asthma App using Spirometer”. For the A3S system development, XAI algorithms will perform patient data analysis and guide decisions. The term “explainable” in XAI refers to the ability of these algorithms to provide transparent and interpretable explanations to patients and health care professionals for their predictions or decisions. The XAI algorithms will provide person-centred interventions precluding the onset or minimizing the risk of asthma in disadvantaged groups. It will help to improve respiratory health by training person-centred AI-support in asthma diagnosis and management of asthma in disadvantaged groups based on social determinants. Overall, the A3S project addresses a pressing health care problem with potentially widespread impact.

Cross-disciplinary collaboration
This is a highly collaborative project with a team of data scientists, clinicians, and AI specialists. Partners are PI Magnus Jansson, Co-PIs Saikat Chatterjee, PostDoc Zhendong Wang of EECS KTH, Ioanna Miliou of DSV SU, Docent Björn Nordlund of KUH and KI and his team, along with MediTuner AB.

About the project

Objective

Background
Wastewater collection networks are essential for ensuring public health and wellbeing, yet they are susceptible to numerous faults including pipe bursts, pump malfunctions, and valve failures. Traditionally, preventing these issues has depended on frequent inspections and reactive repairs. However, there is growing recognition that a more proactive strategy—one rooted in predictive and condition-based maintenance—can both enhance the reliability of wastewater infrastructure and streamline the resources required to operate it. Such an approach can significantly reduce unexpected downtime, extend equipment lifespans, and ultimately lower overall lifecycle costs.

Despite the promise of prognostic models for predictive maintenance in many industries, water infrastructure has not received as much attention as manufacturing or other sectors. Current diagnostic tools in this domain are often tailored to a specific component or pump type, requiring specialized local measurements such as vibrations, oil temperature, or power consumption. In wastewater networks with diverse types of stations and pumps, designing a model for each component can be time-consuming. Moreover, missing measurements or uncertain behavior pose additional challenges. Consequently, the need has emerged for a flexible, data-driven solution capable of handling variations in station design, measurement availability, and environmental conditions.

The DECORUM (Optimized Predictive Maintenance for Wastewater Pump Stations) project, establishes the cooperation between the City of Stockholm, the Stockholm water utility operator (SVOA), the international water technology firm Xylem, and KTH to fill these gaps. SVOA alone operates roughly 300 wastewater pump stations, each with multiple pumps crucial to the city’s sewage system. Through a six-step development plan, SVOA has already taken steps to reduce maintenance costs while preserving high operational reliability. The next milestone is to move from largely manual, reactive procedures toward data-driven, predictive strategies that detect anomalies early, recommend targeted maintenance, and help technicians make informed decisions about when and how to service equipment.

Crossdisciplinary collaboration
The DECORUM project brings together a multidisciplinary team spanning academia, industry, and municipal stakeholders. KTH researchers contribute expertise in systems modeling, predictive algorithms, and robust control, while SVOA provides domain knowledge of large-scale wastewater operations and real-world operational data. Xylem, as a leading water technology company, offers more in-depth insights into cutting-edge pump hardware and software solutions. By uniting these diverse perspectives, the project can address both theoretical and practical challenges, ultimately delivering a flexible, scalable, and impactful predictive maintenance framework for critical urban infrastructure.

About the project

Objective
Towards the digital transformation, this research program aims to establish theory and methods for privacy-preserving localization in wireless networks. A privacy-preserving localization method is supposed to preserve user privacy while enabling communication and localization functionalities.

Background
Telecommunications providers possess vast amounts of wireless connection data that can be used for localization. Users will benefit from this location awareness, gaining access to a plethora of convenient and personalized services that will drive the digital transformation towards more intelligent, economically viable, and socially sustainable societies and industries. The scope of these services spans the gamut from industry 4.0 and home automation to comprehensive healthcare monitoring. However, it is crucial to acknowledge the potential dangers associated with such location-centric data. In the wrong hands, this information could raise privacy concerns, including data misuse, security breaches, and threats to property. In light of these risks, this research program aims to establish theory and methods for privacy-preserving localization in wireless networks.

About the Digital Futures Postdoc Fellow
Hanying Zhao is a postdoc at the Division of Information Science and Engineering at KTH. Her research interests include statistical inference, privacy-preserving technology, and localization.

Main supervisor
Tobias Oechtering, Professor, Division of Information Science and Engineering at KTH

Co-supervisor
Mats Bengtsson, Professor, Division of Information Science and Engineering at KTH