Numerous methods exist for computationally optimising particle simulations, from neighbour identification algorithms like Linked Cells and Verlet Lists to parallelisation schemes and data layouts; however, past works have shown that none of these are optimal for all simulation scenarios, and the best algorithmic configuration can change depending on the particle distribution, interaction model, and hardware [1]. Such findings have led to the development of the particle simulation library AutoPas, which aims to select and configure the most optimal algorithmic configuration for a simulation and change these dynamically throughout a simulation if necessary. Furthermore, it can be used to select different configurations in different regions of the domain to match their distinct characteristics [2]. A range of algorithmic selection methods are available; however, we focus on data-driven methods that use information from the scenario, such as deviations in the distribution of particles, to make their choices. Past works on AutoPas have, however, focused on Molecular Dynamics simulations. In this work, we will discuss the application of AutoPas to Discrete Element Method simulations whilst contextualising our developments with our past research in Molecular Dynamics to provide motivation. These developments include the implementation of a specialist DEM neighbour identification algorithm, hierarchical grids, which can handle different particle scales effectively. We discuss the application of existing AutoPas shared memory schemes to this algorithm and how AutoPas could configure such an algorithm. We will also describe our changes to better handle non-spherical particles. These include extending our automatic Array-of-Structures to Structure-of-Arrays data layout conversion to better SIMD-parallelise such particles. In addition, we will discuss extensions to our algorithm selection methods so that selections can be tailored towards different particle models.