Daniel Watzenig, daniel.watzenignoSpam@tugraz.at

Our research focuses on

• Multi-modal perception and data fusion
• Reinforcement learning
• Data-driven decision making (3DM)

for autonomous systems and autonomous multi-agent systems.

Multi-modal perception and data fusion

• Perception modeling (lidar-radar-camera combinations)
  • AI-based, physical sensor models, hybrid models
  • Sensor fault prediction and detection
  • Dynamic sensing of the vehicle environment considering detection uncertainties
• Fusion of different sensor sources to increase reliability of autonomous systems (probabilistic methods, evidence theory-based, semantics-oriented, etc.)
  • Statistical methods for data association
  • Dynamic occupancy grid filtering
  • Multiple target tracking (MTT) and extended object tracking
    • Multiple detections per object and from multiple targets
    • E.g., Gaussian mixture probability hypothesis density (phd) tracker or Gamma Gaussian inverse Wishart tracker

Reinforcement learning (RL)

• Bayesian modeling and simulation
• Dynamic programming
• Learning from limited samples / sample efficiency
• Interacting with systems that are partially observable
• Learning from multi-objective or poorly specified reward functions
• Stochastic environments
  • Many real-world environments are stochastic, making it difficult for RL agents to learn effectively
• Transfer learning, i.e. transfer of knowledge from one task or environment to another
• RL in path planning of autonomous systems
• RL for multi-agent systems
• Deep learning implementation

Decision making under uncertainty

• (Partial observable) Markov decision processes (POMDP and MDP)
• Modeling of uncertainties in perception and signal propagation (model and state uncertainty)
• Simulation of collaborative autonomous multi-agent systems
• Distributed and collaborative decision making
• Data-driven decision making (3DM) and decision making support capitalizing on AI
• Large language models (LLMs)

.