A key to the performance of a Machine Learning (ML) algorithm is the training dataset for the algorithm as it specifies the desired output of the algorithm and in turn the behavior of the algorithm. The impact of the training dataset becomes more significant when it comes to ML-based online path planning. The training dataset should be created and prepared so that an autonomous vehicle guided by the resulting ML algorithm can be operated safely even in a highly dynamic environment.

To illustrate the impact of the training data on the performance of a ML algorithm for online path planning, we generate different training path data using different path planning algorithms and test the performance of a neural network depending on the training path data. We further examine the properties of the training data that make a ML algorithm more reliable as an online path planner.