Exploring the usefulness of Object Affordances through a Knowledge based Reinforcement Learning Agent
Rupam Bhattacharyya, Shyamanta Moni Hazarika · 2021
The conceptual understanding of human about object usage in executing high level activities is phenomenal. Embodiment of such understanding within a computational agent holds promise in building intelligent household robots for performing such activities. This paper examines the role of object affordances in realizing the high level indoor activities. To this end, we have studied the temporal constraints imposed by object affordances in such temporally extended activities. A novel knowledge based reinforcement learning agent is designed by considering Linear Temporal Logic (LTL). Experimental evaluation demonstrates the usefulness of this agent in performing different activities in grid world environments.