Simulation and Prediction of Human Flows in Hospital Environments from Assistive Robot Observations

This paper introduces a framework for predicting individual motion patterns in hospitals by extending the Flowgrid model to account for specific agent roles and temporal rhythms. Utilizing the BePed simulator, the system enables multi-robot teams to navigate socially by distinguishing between the movement behaviors of doctors, nurses, and patients.

Predicting human locations and flows is a key factor in social robotic navigation. This paper presents a framework for modeling and predicting individuals’ motion patterns from local robot observations in hospital environments. We extend the human flow grid (Flowgrid) model to distinguish between agent categories (e.g., doctors, nurses, and patients) and propagate their presence probabilities using typed Flowgrids1. Time-aware Flowgrid management further enables adaptation to the temporal rhythms of hospital operations. The current work is illustrated with the help of a pedestrian simulator called BePed. The framework is demonstrated in a simulated hospital environment with a multi-robot team. This work is part of the SOLAR-NAV project.