D3.3 - Radiation mapping

There are obvious challenges to surveying an outdoor site remotely to reveal radiation sources and localise them autonomously. This project set out to improve existing solutions. The team used a wheeled robot that could autonomously traverse challenging terrains. A probabilistic radiation map was incrementally built, and sensing points were automatically chosen to drive the autonomous exploration of the robot.

A hybrid digital twin-real-world system was deployed to facilitate experimentation with simulated radiation sources in controlled positions, while researchers tested navigation and movement behaviours in the real world under realistic conditions.

Robot navigation was performed through the use of a topological map: a graph-based representation that exploits the structure of the environment. The robot path planning was performed by Next-Best-Sense, a framework for combining multiple criteria and expressing the preference of one above others. Kriging method (also known as Gaussian Process regressor was used for interpolating sensor readings and building two maps (mean and covariance of where the radiation sources were located