The paper presents a new trajectory prediction model, Social Force Generative Adversarial Network (SoFGAN), for human-robot collaboration. It uses a Generative Adversarial Network (GAN) and Social Force Model (SFM) to generate different plausible people trajectories reducing collisions in a scene. A Conditional Variational Autoencoder (CVAE) is added to emphasize the destination learning. The model is more accurate in making predictions and also reduces collisions compared to other approaches. The model can be used in real-time without GPUs to perform good quality predictions with a low computational cost.
Publication date: 17 Nov 2023
Project Page: https://arxiv.org/abs/2311.10582v1
Paper: https://arxiv.org/pdf/2311.10582
