Radioactive contamination monitoring is an important part of radiological protection. Automation of surface monitoring poses difficulties, with a major challenge being determining the coverage of a radiation probe over an object in close proximity without collision. We propose a new accessibility framework to determine if radiation probes, modelled as convex hulls, collide with 3D point clouds representing the objects. We explain how to structure and analyze point clouds to extract properties such as the surface normal for each point. Our method for approximating radiation probes is demonstrated using the BP4 probe. This approach models both the probe and the sensor's effective scan volume with geometric primitives, providing a computationally efficient way to detect collisions with flat surfaces. The accessibility assessment builds on common methods within computer science for determining intersections. A laptop in various positions was used to demonstrate that the framework can efficiently categorise points as accessible or inaccessible, identifying unscannable regions. The output of this framework can then be used to plan collision-free paths over objects and will be the foundation of a robotic survey assistant.
Many industries have confined, clustered environments that require routine inspections. Nuclear facilities, in particular, have extensive pipe networks that pose additional radiological challenges, making human inspection unsuitable. A mobile manipulator can inspect areas unreachable by most mobile robots without the additional risks and hazards associated with using aerial robots. This work formulates a whole-body controller for a non-holonomic mobile manipulator as a Quadratic Program, utilising Vector Field Inequalities (VFIs) to constrain the system. We introduce a stochastic point-to-plane constraint to handle uncertainty in the walls’ localisation and an online path planner that combines coverage path planning and 3D reconstruction. We present our recent results on coverage path planning and discuss the process of redefining the task to combine it with the NBV into a single approach. We demonstrate how to extract constraints from point clouds and incorporate their uncertainties into the VFI framework, and we comment on the performance of these constraints at preventing violations. We demonstrate this through experimental results from a physical mobile manipulator operating in a confined environment, thereby proving the real-world viability of these approaches and concluding with a discussion of future work to generalise them to other primitives and environments and to provide formal guarantees.