Developing an AI/optimisation approach for the autonomous dismantling and packing of legacy nuclear installations

Lead Research Organisation: University of Leeds
Department Name: Chemical and Process Engineering


At some stage in the decommissioning of nuclear installations, it is inevitable that large metal structures (e.g., reactor vessels, gloveboxes, structural components) need to be cut into smaller pieces to be packed in containers, for temporary storage, permanent disposal or simply for transportation. Both cutting and packing cost money and usually, efficiency or cost-saving in one is achieved at the expense of the other (meaning a trade-off is always required). For example, consider a case where a radiation-protected human operator goes into a hot cell to take down an installation. Because of radiation exposure limits, there is usually not much time for deliberation. As such, the decision on how best to approach the problem is typically made there and then at the discretion of the human operator, and once a cut is made, there is no going back (i.e. there is very little margin for error). Due to scenarios like these, there is ongoing interest in the development of robotic/AI technologies for carrying out nuclear decommissioning tasks such as structure scanning, radiation mapping, and the dismantling and packaging of waste. The use of robotics/AI for such tasks can in principle remove constraints relating to radiation exposure time (hence allowing for longer operating times), whilst also allowing for more cost-efficient operation. This project is motivated by the need for more cost-effective ways to cut and package nuclear waste during decommissioning tasks. Currently, cutting and packing software can be used to perform cutting and packing trials entirely on a computer, but the problem with existing approaches is that they perform the cutting and packing tasks independently. This means that, for every set of objects generated by a cutting algorithm, a packing algorithm must then be used to trial different packing combinations for each set of objects. As such, if a human operator wishes to find the best way to cut an object so that the maximum amount of cut parts can be packed into a container, they will have to use a brute force approach to trial every possible cutting and packing combination until the best solution is found. This approach is problematic as even for a simple cutting/packing task the number of cutting/packing trials would quickly spiral out of control, resulting in excessively large computation time.

The aim of this project is to work out and introduce an artificial intelligence approach to link the two processes so that the following goals can be achieved:

Minimise total costs: the primary goal of this project is to create an algorithm which can minimise the total cost of the cutting/packing process.

Reduce number of cutting/packing trials: as stated, one of the problems with cutting/packing tasks is large the number of possible combinations. As such, a more intelligent approach is required to search for a more promising subset of combinations to trial rather than doing a brute force search over every possible combination.


Short of brute force trials, there is an intelligent and scientific way to optimise, under

various constraints, the cutting/packing process for overall cost saving, assuming accurate structural models and radiation maps are available (from in-situ, real-time, robotically operated scanning).

High Level Objectives:

Search and review relevant AI/optimization routes.

Develop them such that machine autonomous operation and decision-making is possible.

Demonstrate feasibility through simulations.

If possible, implement and demonstrate at lab scale using real robotics.


10 25 50

Studentship Projects

Project Reference Relationship Related To Start End Student Name
EP/S022295/1 31/03/2019 29/09/2027
2439552 Studentship EP/S022295/1 30/09/2020 29/09/2024 Aron Oliver Webster