Mechanical weed management is a major bottleneck for sustainable yet efficient food production. Although inter-row mechanical weeding has been used for many years, intelligent and versatile intra-row weeding solutions remain unavailable due to the high complexity of agricultural environments, which show a great deal of variability. However, recent developments in both robotics hardware and machine learning, put this task within reach of automation. This project aims to create a learning methodology that combines model-based learning of dynamics based on multi-modal sensory input with model-free residual reinforcement learning in a simulated environment to learn adaptive control policies for weed management with a mobile robot platform in an agricultural setting. The research will be conducted in collaboration with IDLab-AIRO at Ghent University and the Flanders Research Institute for Agriculture, Fisheries and Food (ILVO). It builds upon an agricultural robot platform developed within this partnership. The research outcomes of this project will be of direct impact on both the Flemish primary and industry sector.