Mobil Robotlarda Sosyal Navigasyon Öğrenme Learning Social Navigation in Mobile Robots

Traditional approaches to robot navigation have prioritized physical aspects such as efficiency and collision avoidance. However, the social competency of robots has a profound effect on our acceptance of them in our daily lives. Current techniques for improving social compliance have relied on assumptions about social behavior and predefined features or reward functions. To overcome this limitation, we propose a new Learning from Demonstration framework that uses the position data generated by a supervisor. The model we propose models the spatial dimension of the target social behavior with Graph Attention Networks and the temporal dimension with Conditional Neural Processes. The performance of the proposed model is eva- luated quantitatively and qualitatively using simulated scenarios with different numbers of pedestrians.