Embodied AI for dexterity-capable construction Robots: DEXBOT framework

Abstract

The construction industry is integrating robots into critical tasks at an accelerated pace, with the aim of enhancing efficiency, safety, and productivity. However, construction tasks requiring dexterity remain a challenge due to the need for precise movements, accurate perception, real-time decision-making, and a comprehensive understanding of the environment. To address these challenges, the introduction of embodied artificial intelligence (AI) represents a significant shift in robotic capabilities to enhance their alignment with the broader spectrum of construction settings. Rooted in cognitive science, embodied AI emphasizes the integration of an agent’s physical form into its computational intelligence processes. It resembles how humans develop motor skills by interacting with physical world. This paper introduces DEXBOT, an exploratory framework for designing construction robots capable of high dexterity using embodied AI principles that mimics human strategies in complex tasks. The framework outlines six key perspectives for solving high-dexterity tasks with embodied AI: scene understanding, localization and motion planning, position-based control, force-based control, sequence planning, and correction decision-making. By presenting preliminary test cases for each perspective, the paper emphasizes the role of embodied AI in advancing dexterity level of construction robots. The DEXBOT framework is expected to encourage interdisciplinary collaboration of designing capable construction robots in the future.