Simulating Pedestrian Flow on Slopes via Transfer Learning Approach: From Single-File to Crowd

The increased prevalence of stairs and ramps in urban areas has highlighted the importance of understanding pedestrian dynamics on inclined surfaces. Previous studies indicate that pedestrians exhibit distinct motion characteristics on slopes compared to level ground. While traditional rule-based models used for pedestrian simulation may not accurately capture the complexities of pedestrian behaviour on slopes, this paper presents a Transferable Pedestrian Motion Simulation Network (TPMSN) tailored for accurately modelling pedestrian dynamics on slopes. The TPMSN incorporates five key input features to capture relative position information, neighbour motion states, and trends in pedestrian motion. Through a transferable layer, the network demonstrates versatility in handling crowd simulation tasks across different slope angles. Networks pre-trained with data from pedestrian single-file flow experiments exhibit robust fitting performance, with R-square values ranging from 0.848 to 0.999. Furthermore, transfer learning is applied with experimental data from pedestrian unidirectional flow on slopes. Two independent networks are trained to predict instantaneous velocity vectors. Simulations carried out by the two networks demonstrate promising accuracy and authenticity, which are validated by a mean Average Displacement Error (ATE) of 0.0926 m and reproduction of flow-density fundamental diagrams. Additionally, the networks could capture pedestrian lateral body sway, a crucial aspect of real-life pedestrian behaviour on slopes, as evidenced by lane entropy trends consistent with empirical studies. Overall, the TPMSN offers a successful approach for crowd simulation on slopes. This work contributes to the advancement of crowd simulation techniques in complex terrains, offering valuable implications for urban planning, crowd management and architectural design.