Human Movement Prediction is Shared Human-Robot Workspaces

Robots and humans increasingly coexist in shared spaces. Manufacturing workspaces pose additional requirements as fleets of mobile vehicles are deployed alongside workers to jointly or independently perform activities. Breaking the clear separating zones provisioned in the past for automated guided vehicles and humans, safety and legal concerns must be addressed to make shared industrial workspaces both safe and productive. This creates a need for research in socially-aware robot navigation to target also such deployments. To allow for safer and more efficient manufacturing operations in shared workspaces, it is important for robot fleet planning to anticipate human movement. The research related to the datasets here aims to explore and evaluate the extent to which a graph-based neural network approach can be applicable to predicting human occupancy in such workspaces based on past observations. The approach applies a spatiotemporal graph neural network that uses graph convolution and gated recurrent units, and an attention mechanism to capture the spatial and temporal dependencies in the data. Addition of learnable edge weights, a feature exclusive to graph neural networks, enhances the predictive capabilities of the model. Adding workspace – specific embeddings to graph nodes has also been explored, bringing modest performance improvements. The original dataset has been produced by THALES Six as part of the Horizon 2020 EU project STAR (<a href="https://star-ai.eu/">star-ai.eu</a>), grant ID 956573.