EvacAware⁚ Real-Time Intelligence for Coordinated Evacuations

This project aims to design a practical evacuation assistance system for evacuation leaders and teams to improve their situation awareness in emergencies through a mobile application.

Evacuating buildings during emergencies such as fires or terrorist attacks requires heightened environmental awareness, quick decision-making, and immediate action. This project proposes EvacAware, a practical evacuation assistance system designed to enhance situation awareness for evacuation teams via a mobile application. With EvacAware, team leaders gain real-time awareness of critical factors, such as environmental hazards, evacuation procedures, and requests for additional assistance.

EvacAware primarily aims to enhance the situational awareness of evacuation teams, composed of multiple leaders and support a coordinated evacuation. To achieve this, a team situation awareness module is integrated into the system. Additionally, a comprehensive sensory infrastructure is proposed to collect real-time environmental data, recognize situational contexts, and determine users’ indoor positions within the building. Advanced data analysis techniques are employed to predict situations, assess risks, and improve overall understanding of both the environment and user conditions. This enables the system to provide the evacuation team with real-time environmental insights and actionable recommendations.

Both evacuation leaders (teams) and building occupants can use the application. Occupants can inform evacuation leaders about their status and, when necessary, be tracked by the leaders. The mobile app is designed to minimize information overload and maintain a user-friendly interface, even in high-stress situations.

The app focuses on evacuating large buildings with numerous occupants, such as schools, business centers, hospitals, airports, train stations, and shopping malls, during emergencies. EvacAware can function as a comprehensive system across different buildings, allowing users to operate a single mobile application for all EvacAware-integrated facilities.

• Date:
• Contributors: A. Sarkheyli-Hägele, J. Holmberg, R. Malekian, J. Persson, G. Hägele, E. Akin, D. Surie, A. Bouguerra
• Funded: The Crafoord foundation
• External links: MAU Project page
• Related Publication:

Activity-Aware Smart Environments

Machine Learning and IoT Sensors for Human Activity Recognition

The project aims to explore the use of machine learning and IoT technologies to recognize and analyze human activities in real time. By leveraging ambient IoT devices such as wearable and portable devices, radar, camera systems, and advanced AI models, the project seeks to develop intelligent systems that can support applications such as health monitoring, smart home automation, workplace safety, and personalized assistance. This research will provide actionable insights into human behavior, enable proactive interventions, and contribute to the design of more responsive, interactive, and adaptive environments.

Trustworthy LLM-Based Decision Support for Sustainable Mobility

The project investigates how Large Language Models (LLMs) can be applied to analyse travel behaviour patterns among specific social groups. It also evaluates the impact of interventions and policy changes on mobility. While testing a proof of concept, the project aims to deliver actionable insights that inform policy and support more sustainable transport systems.

Federated and Privacy-Preserving Machine Learning

The project aims to investigate and develop federated and privacy-preserving machine learning methods that allow multiple devices or organizations to collaboratively train AI models without sharing raw data. By leveraging techniques such as federated learning, secure aggregation, differential privacy, and edge AI, the project seeks to support applications in areas such as smart environments, healthcare, mobility, public safety, and organizational analytics. The ultimate goal is to enable trustworthy, scalable, and privacy-aware AI systems that respect data ownership, regulatory constraints, and ethical principles.

Situation Awareness-based Attention Guidance

Situation Awareness-based Attention Guidance (SAAG) is aimed at developing a novel Situation Awareness component (SAc) for an Attention Guidance system to improve human situation awareness in complex environments and tasks.

Situation Awareness-based Attention Guidance (SAAG) aims to develop a novel Situation Awareness component (SAc) for an Attention Guidance system to improve human situation awareness in complex environments and tasks. The Situation Awareness-based Attention Guidance (SAAG) system, focusing on SAc, is realized and evaluated for the players of an Augmented Reality (AR) tabletop game. Situation awareness is crucial for decision-making and action-taking.

The awareness emerges from relating knowledge about the environment, human goals, and objectives. Accordingly, the SAc will mimic this process artificially through a novel integration of four functional modules responsible for perception, situation recognition, intention recognition, and projection. The SAc outputs a situation-action space considering the user preferences, goals, and intentions used as input for decision-making.

The project aims for both short-term scientific and long-term industrial and societal impact. It is revealed as a human assistance system prototype for complex situations. In addition, it has concrete contributions to artificial situation awareness to be published in academic venues covering human-computer interaction, situation awareness, and AR games. The most immediate societal impact of the developed SAc will probably be embedded in persuasive technologies for reaching long-term behavioral change for individuals and society in the areas of healthcare, safety, and climate change.

• Date:
• Contributors: A. Sarkheyli-Hägele, J. Font, D. Kadish, T. Pederson, G. Hägele
• Funded: The Crafoord foundation
• External links: Video, MAU Game Lab, MAU Project page
• Related Publication:

Safety Assurance of ML-Based Algorithms

Mitigating Risks and Ensuring Reliability in AI System

This project is dedicated to tackling the crucial challenge of guaranteeing the safety and reliability of machine learning (ML) algorithms across diverse applications. With ML algorithms becoming integral to sectors like healthcare, machinery, and transportation, ensuring their safety is essential for widespread adoption, trust, and legal compliance.

Our research delves into advanced methodologies, including uncertainty estimation and conformal prediction, to enhance ML algorithms’ monitoring and runtime verification. We expect continuous monitoring of algorithmic performance and behavior to enable the detection of anomalies and safety-critical deviations. This proactive approach will allow timely interventions to uphold systems’ safety and reliability, thus bolstering trust in ML systems and their use within safety-critical applications.

• Date: 2023 - ongiong
• Contributors: G. Hägele, A. Sarkheyli-Hägele, A. Bouguerra
• Related Publication:

Supervisor-based Safe Behavior Assurance

Supevirory safety unit to ensure overall system's safe behavior

Autonomous or semi-autonomous systems are used in different application domains to simplify or assist human tasks. In this context, their behavior has to be verifiably safe. This project focuses on techniques for specifying and designing a supervisory safety unit to ensure the overall system’s safe behavior. This includes methods for run-time risk area generation and adjustment, situational hazard analysis and risk assessment, run-time technical system reliability assessment, and emergency behavior modeling are investigated.

• Date: 2013 - 2017
• Contributors: G. Hägele
• Related Publication:

Understanding Implicit Object Relationships for Enhanced Situation Awareness in AI Systems

Unveiling Subtle Connections to Elevate AI Perception and Response

In this dedicated project, our aim is to bolster artificial situation awareness by delving into the realm of implicit object relationships. While explicit relationships between objects are often readily discernible, it is the implicit connections that often hold the key to a deeper understanding of complex situations. Our research focuses on developing algorithms and methodologies to uncover these subtle links, thereby enriching the perception capabilities of AI systems.

Using a combination of machine learning techniques, data analysis, and contextual understanding, we embark on a comprehensive investigation.

• Date: 2024 - ongiong
• Contributors: A. Sarkheyli-Hägele, G. Hägele
• Related Publication:

Situational Risk Assessment

Situational risk assessment in open and non-predictive environments to support decision making and ...

… action selection for autonomous systems is a challanging task and focus of this project. Decision making and action selection incorporate decreasing situational risks and maintain safety as operational constraints. Commonly, neither existing safety standards nor the situation modeling or knowledge representation is considered in that context.

This project investigates and applies novel approaches related to Safety-Driven Behavior Management for different applications like unmanned aerial systes or automated driving. The focus lies mainly on situation modeling, and the problem of knowledge representation in the context of situational risks. One of the research questions is whether the safety standards-oriented hazards analysis and the risk assessment approach can be combined with the machine learning-based situation recognition.

• Date:
• Contributors: G. Hägele, A. Sarkheyli-Hägele
• Related Publication: