This article presents HOI4ABOT, a framework that allows robots to anticipate human-object interactions (HOIs) and understand human intentions for better collaboration. The proposed transformer-based model can efficiently detect and anticipate HOIs from videos, enabling robots to assist humans more proactively and intuitively. The model outperforms current state-of-the-art results in HOI detection and anticipation while being significantly faster. This approach has been implemented on a real robot, demonstrating improved human-robot interaction.
Publication date: 28 Sep 2023
Project Page: evm7.github.io/HOI4ABOT
Paper: https://arxiv.org/pdf/2309.16524
