Ömer Erduran

Ömer Erduran

Research Assistant, Computer Science, Goethe University Frankfurt

Projects

Here’s a selection of my research projects, primarily focused on intelligent agents, mobility simulation, and machine learning.

Cognitive BDI Agents for Ride-Hailing

  • Description: Developed BDI agents that negotiate shared ride offers using realistic decision models. The agents are integrated with MATSim for simulation in large-scale urban scenarios.
  • Technologies: Jadex BDI, Java, MATSim
  • Highlights:
    • Implemented trip negotiation logic.
    • Integrated reasoning-based agents into a transport simulator.
  • Paper: ICAART 2022
  • Code: GitHub Repository

Cluster-Based Vehicle Placement for Ride-Hailing

  • Description: Applied unsupervised learning to analyze ride request patterns and optimize vehicle placement in urban areas.
  • Technologies: Python, scikit-learn, NumPy, Pandas, MATSim
  • Highlights:
    • Spatial clustering of trip origins and destinations.
    • Evaluation in synthetic and real-world traffic datasets.
  • Paper: ICMLA 2019
  • Code: GitHub Repository

Offline MARL for Shared Mobility

  • Description: Investigated offline reinforcement learning techniques for multi-agent coordination in ride-pooling environments.
  • Technologies: PyTorch, RLLib, Offline RL, MATSim, DJL
  • Highlights:
    • Custom offline datasets generated from MATSim.
    • Evaluation of multi-agent performance under varying fleet sizes.
  • Paper:
  • Code: GitHub Repository

Deep Reasoning in BDI Agents

  • Description: Conceptual exploration of integrating deep learning with symbolic reasoning inside BDI architectures.
  • Technologies: Jadex, PyTorch, Planning Logic
  • Highlights:
    • Surveyed hybrid neuro-symbolic architectures.
    • Proposed a modular integration strategy for ML in agent decision-making.
  • Paper: Erduran. “Deep Reinforcement Learning for Software Agents in Mobility on Demand”. LWDA 2024 Workshop: FGWM 2024.
  • Code: GitHub Repository

MATSim-Jadex Integration Framework

  • Description: Built a software bridge between MATSim and Jadex to enable simulation of cognitively rich agents in realistic traffic environments.
  • Technologies: Java, MATSim, Jadex
  • Highlights:
    • Generic interface for perception-action loops.
    • Easily extensible for new agent behaviors.
  • Paper: EMAS 2024
  • Code: GitHub Repository

🔗 For more, check out my GitHub or reach out on LinkedIn!