Internship | Visibility-Aware Counterfactual Motion Planning Using a Learned World Model

About this position

In this assignment, the student will design a system that uses a vision-based world model to predict whether a robot would be visible to another agent after executing a small movement (e.g., peeking around a corner). From monocular or RGB-D observations, the system should reconstruct approximate 3D scene geometry, estimate the other agent’s pose, and simulate counterfactual robot poses. By performing geometric line-of-sight and field-of-view reasoning from the other agent’s viewpoint, the system will compute a visibility score for candidate movements and select actions that minimize exposure. The project combines scene reconstruction, pose estimation, and visibility-aware planning, and frames safe navigation as a counterfactual reasoning problem over a learned world model.

What will be your role?

You will perform this assignment within TNO’s Intelligent Imaging department. The Intelligent Imaging department is a pa...