Proceedings of the Thirtieth International Joint Conference on Artificial Intelligence (IJCAI-21) Survey Track Building Affordance Relations for Robotic Agents - A Review Paola Ardon´ , Eric` Pairet , Katrin S. Lohan , Subramanian Ramamoorthy and Ronald P. A. Petrick Edinburgh Centre for Robotics Edinburgh, Scotland, United Kingdom
[email protected] Abstract practical implementations of affordances, with applications to tasks such as action prediction, navigation and manipulation. Affordances describe the possibilities for an agent The idea of affordances has been studied from different to perform actions with an object. While the sig- perspectives. Early surveys [Chemero and Turvey, 2007; nificance of the affordance concept has been previ- S¸ahin et al., 2007; Horton et al., 2012] summarise formalisms ously studied from varied perspectives, such as psy- that attempt to bridge the controversial concept of affordances chology and cognitive science, these approaches in psychology with mathematical representations. Other sur- are not always sufficient to enable direct transfer, veys discuss the connection of robotic affordances with other in the sense of implementations, to artificial intel- disciplines [Jamone et al., 2016], and propose classification ligence (AI)-based systems and robotics. However, schemes to review and categorise the related literature [Min many efforts have been made to pragmatically em- et al., 2016; Zech et al., 2017]. In contrast, we focus more ploy the concept of affordances, as it represents on the implications of different design decisions regarding great potential for AI agents to effectively bridge task abstraction and learning techniques that could scale up perception to action. In this survey, we review and in physical domains, to address the need for generalisation in find common ground amongst different strategies the AI sense.