Collaborative Perception and Learning

In the Internet of Things, a heterogeneous multitude of smart devices with multi-modal sensors will collaborate to form a perception of their environment, realizing the concept of “ubiquitous sensing”. In order to solve the addressed application use cases, they will employ data-driven decision-making, based on machine learning algorithms. The key question is how to match an edge computing architecture suitable for AI with an AI architecture suitable for edge computing. Paradigms of the future will combine data-driven, learning approaches with model-based, symbolic AI, while using knowledge exchange in the network of agents to solve the trade-offs imposed by resource constraints and dependability requirements.

Research Foci

  • Multi-agent reinforcement learning
    • Solving inherently distributed use cases in the interaction of the agent with its environment
  • Explainable AI
    • Hierarchical models, ambiguity and multiple dimensions of classification
  • Computer vision (CV) meets sensing
    • Using CV to create generic sensor replacements
    • Using CV techniques and algorithms for non-visual sensor modalities
  • Anomaly detection, industrial process optimization and predictive maintenance
  • Sensor fusion, virtual sensing, ubiquitous sensing concepts

Research Competencies

  • Machine learning techniques, specifically reinforcement learning
  • Computer vision algorithms, object recognition and tracking
  • Time series data analysis, statistical evaluations, virtual sensing & anomaly detection
  • Fog and edge computing, distributed systems, distributed microservices architectures


  • Anomaly detection and degradation tracking in multi-modal data monitoring photovoltaic plants
  • Wear estimation of mechanical parts of an industrial machines through virtual sensing
  • Volume estimation of objects using a series of 2D images and camera movement
  • Multi-dimensional object classification in images to enhance explainability

Your contact person

DI Dr. techn. Willibald Krenn

Head of Research Unit Trustworthy Adaptive Computing


Member Area