My first step, is to build hardware and software systems simultaneously to control the AUV. An important consideration when planning the construction of the AUV would be to construct a form that can easily and effectively be modelled using mathematics to reduce computations required in the simulation thus greatly reducing the time required and the recurrence of errors. An another key factor would be to ensure that the robot uses hardware and software systems that establish accurate and efficient control of the robot. This allows for the transferability of software between the robot's internal algorithms and simulation for control over thrusters and increased accuracy from the modelling. The AUV would also need to be able to accurately measure its position in space. Thus it would require the implementation of various sensors and potentially processing algorithms in order to enable comparison with its simulated counterpart.

The first challenge that serves as a prerequisite for the experiment would be to first find a mathematical model for the simulation that can match the movement of the AUV. It would be required that the state of the robot in the simulated and real world roughly match each other in minimal test. If the mathematical models are inconsistent at this level, it would be meaningless to proceed to the next step of training the AI as the model has already been invalidated as a candidate for accurately simulating the real world.