Many believe that to achieve artificial general intelligence (AGI), we must endow agents with common sense, including an understanding of object physics (e.g. containers, permanence, enclosures). Shanahan et al. explain how the same experiments used to study animal cognition (including understanding of objects, space, and causality) can be applied to train and evaluate how well an RL agent in a 3D environment understands objects and their affordances. They also argue that research on animal cognition can inform the design of new RL architectures and experiments.