Robots that share walkways with people need to anticipate where pedestrians are heading, but standard semantic occupancy maps only capture the present. A new arXiv preprint introduces ReForesight, an open-source ROS 2 toolkit designed to fill that gap by adding instance-level future occupancy and re-identification to dynamic semantic mapping.

The toolkit tracks individual people over time and predicts where they will be in the near future, allowing robots to plan paths that avoid likely collisions. By re-identifying instances across frames, it maintains a consistent understanding of each pedestrian rather than treating them as anonymous obstacles.

Because ReForesight is released as open-source software for ROS 2, it can be integrated into existing robot stacks for research and deployment. The authors position it as a practical step toward safer and more socially aware navigation in environments where robots and humans routinely cross paths.