<p>Practical deployments of exoskeletons can often be limited by cost, limiting access to their usage by those that would benefit from them. Minimising cost whilst not harming effectiveness is therefore desirable for exoskeleton development. For Control Systems governing assistive and rehabilitative exoskeletons that react to the wearer’s movements, there will inevitably be some delay between when their wearer intends to move and when the exoskeleton can assist with this movement. This can lead to situations where a user may be limited by their own assistive exoskeleton, reducing their ability to move freely. A potential solution to this is to provide a proactive method of control, where the most likely path of the wearer’s movement is predicted ahead of the wearer making the motion themselves. This can be used to give the user assistance immediately as they are walking, as well as potentially pre-emptively adjust their gait if they suffer from predictable gait deficiencies. The purpose of this paper is to investigate the Data Collection, Implementation, and Effectiveness of an LSTM Recurrent Neural Network dynamically predicting future movement based off of prior movement. These methods were developed to use off the shelf, Low-Cost Microcontrollers as to minimise their Financial, Weight, and Power Impact on an overall Low-Cost exoskeleton design, as well as to evaluate how effective such an implementation would be when compared to running such a Neural Network on a more powerful processor. The created model was capable of achieving similar accuracies to far more powerful models on High-Powered Laptops.</p>