This agent-based model was developed using the Unity game engine to incorporate multi-agent reinforcement learning algorithms from the ml-agents OpenAI package. The model simulates offender agents over a 2D landscape containing interventions, targets, and routine activity nodes. Offenders train using a multi-agent reinforcement learning algorithm Proximal-Policy Optimisation (PPO) to learn behaviours that demonstrate realistic patterns of burglary in agreement with the Rational Choice Perspective, Crime Pattern Theory and Routine Activity Theory. The novelty presented by this model is based on the ability for offender agents to learn behaviours naturally from the environment without any hard-coded pre-determined behavioural rules. Users can test Situational Crime Prevention Intervention (SCPI) policies where interventions can be placed in a specific location run-time, thus, increasing risk in the area and the reactions of offenders can be analysed. Overall, the experiment results show that offenders learn to offend at targets where rewards outweigh risks and effort, demonstrating a degree of intelligence, such as offending closer to home, frequently victimising high-rewarding targets, and learning to avoid areas of high risk. Read the CrimeABMDocumentation.pdf document which outlines each step to run the simulation, including steps to train your own RL offender agents in a specific situational crime prevention intervention scenario. The work uploaded here is part of an ongoing research project which will be published as a research article.