Task Sensing and Adaptive Control for Mobile Manipulator in Indoor Painting Application

Robotic painting, particularly in industrial and construction domains, has attracted considerable attention due to its precision and uniformity. However, current systems are constrained by inadequate precision and effectiveness in painting, particularly when applied to large-scale surfaces. This article introduces an advanced adaptive robotic painting system that incorporates a mobile manipulator (MM) designed to enhance both accuracy and efficiency in indoor surface painting through two innovative submodules: automated trajectory generation and MM adaptive control policy (ACP). Initially, to autonomously generate the accurate trajectory, we propose the attention-aware graph network for refining 3-D surface model to significantly enhance the accuracy and efficiency of environment modeling. Following this, the RayCast 3-D mapping technique is introduced for precise projection of 2-D images onto arbitrary 3-D surfaces with its flexibility and adaptability. Furthermore, we introduce an MM ACP comprising a trajectory controller and a close-loop whole-body controller. This dual-controller system enables the MM to swiftly move to target poses and smoothly follow trajectories, with the capability to autonomously switch between control paradigms based on task requirements. In addition, Experimental results demonstrate that the proposed automated trajectory generation strategy, coupled with the MM ACP, significantly improves the accuracy of environmental perception and the efficiency of trajectory generation. Furthermore, the MM exhibits robust performance in both simulated and real-world settings, successfully executing fully autonomous painting tasks.