Abstract
As a popular concept proposed in the field of psychology, affordance has beenregarded as one of the important abilities that enable humans to understand andinteract with the environment. Briefly, it captures the possibilities andeffects of the actions of an agent applied to a specific object or, moregenerally, a part of the environment. This paper provides a short review of therecent developments of deep robotic affordance learning (DRAL), which aims todevelop data-driven methods that use the concept of affordance to aid inrobotic tasks. We first classify these papers from a reinforcement learning(RL) perspective, and draw connections between RL and affordances. Thetechnical details of each category are discussed and their limitationsidentified. We further summarise them and identify future challenges from theaspects of observations, actions, affordance representation, data-collectionand real-world deployment. A final remark is given at the end to propose apromising future direction of the RL-based affordance definition to include thepredictions of arbitrary action consequences.