Back to projects

2025 / Undergraduate research assistant

Learning Social Navigation in Mobile Robots

A robotics research project on learning socially compliant navigation behaviors in dynamic pedestrian environments.

  • Python
  • PyTorch
  • Unity
  • Graph Attention Networks
  • Conditional Neural Processes

At Boğaziçi University COLORSLAB, I worked on a social navigation project for mobile robots. The work was published as Mobil Robotlarda Sosyal Navigasyon Öğrenme / Learning Social Navigation in Mobile Robots at IEEE SIU 2025.

The project focused on navigation behaviors that are not only collision-free, but socially legible: for example, passing behind pedestrians rather than cutting through their likely path. We modeled the environment as a dynamic social graph and combined graph-based interaction modeling with temporal trajectory prediction.

My Contributions

  • Designed and implemented a hybrid learning framework using Graph Attention Networks and Conditional Neural Processes.
  • Reverse-engineered and extended the Unity-based SEAN 2.0 simulation environment for custom data collection.
  • Built dynamic multi-agent scenarios for training and evaluation.
  • Validated the learned model in social environments against simpler linear baselines.

Outcome

The work produced a conference paper and a research prototype that reduced social norm violations compared with baseline navigation behavior in the evaluated scenarios.