V. SUMMARY The main objective of this thesis was to compare artificial intelligence algorithms for collision avoidance in the field of collaborative robots. Moreover, the main task was to identify algorithms that are used for collision avoidance and use these methods as a part of a neural network, to test and understand which algorithm is the most efficient, accurate, and fastest to incorporate learning. Apart from the algorithms we dived into the implementing a neutral neural network architecture which doesn’t complement any algorithm. Benefit of using the proposed Feed-Forward Neural Network is that learning happens unidirectionally and allows the algorithms themselves to make their own modifications in order to obtain the best-predicted results. Apart from this, the main challenge was to design a couple of digit fields where the algorithm can be tested for the collaborative robot. 4 environments were created, and every single environment had different amounts of dynamic and static objects. The environments surrounding from 1-4 were built to test the KUKA LBR4+. The pretext under the design of these environments was that the difficulty should arise from the first to the last one. The 3rd simulation environment consisted of the most objects, and this resulted in showcasing unexpected results. Conjointly, environment 4 was built to be the most unpredictable, trying to replicate a real-life scenario. Moreover, all four, Levenberg-Marquardt, Resilient Backpropagation, Scaled Conjugate Gradient, and Bayesian Regularisation were simulated in all the environments. A total of 64 simulations were conducted, the breakdown of the 64 is as follows; 4 environments to test, 4 algorithms proposed, and repeating the simulation 4 times to obtain an average and observe more accurate results. The creation of a neural network to implement an AI solution was the main purpose of this thesis.