Human trajectory prediction and generation using LSTM models and GANs

Human trajectory prediction is an important topic in several application domains, ranging from self-driving cars to environment design and planning, from socially-aware robots to intelligent tracking systems. This complex subject comes with different challenges, such as human-space interaction, human-human interaction, multimodality, and generalizability. Currently, these challenges, especially generalizability, have not been completely explored by state-of-the-art works. This work attempts to fill this gap by proposing and defining new methods and metrics to help understand trajectories. In particular, new deep learning models based on Long Short-Term Memory and Generative Adversarial Network architectures are used in both unimodal and multimodal contexts. These approaches are evaluated with new error metrics, which normalize some biases in standard metrics. Tests have been assessed using newly collected datasets characterized by a higher diversity and lower linearity than those used in state-of-the-art works. The results prove that the proposed models and datasets are comparable to and yield better generalizability than state-of-the-art works. Moreover, we also prove that our datasets better represent multimodal scenarios (allowing for multiple possible behaviors) and that human trajectories are moderately influenced by their spatial region and slightly influenced by their date and time