Geometry and Force Guided Robotic Assembly With Large Initial Deviations for Electrical Connectors

Abstract

Electrical connectors (ECs) are extensively employed in industrial scenarios, and their assembly quality is crucial. However, these connectors are often located in confined spaces, which poses challenges of complex lighting conditions and visual occlusions during the execution of robotic assembly tasks. Hence, guiding robots to assemble solely through force/torque (F/T) feedback is an alternative way. However, there is currently limited research on achieving assembly tasks solely through F/T information, especially when the spatial pose of the socket is uncertain, and how to achieve the robotic assembly of ECs remains a difficult problem. To this end, this paper proposes a novel strategy for assembling ECs under large initial pose deviations. Specifically, inspired by observing the assembly process of humans without visual assistance, the robotic assembly is first divided into two stages: by arbitrary surface tracking, the relative pose of the current EC is confirmed and adjusted to a dual-point contact state, i.e. the conversion from non-contact to dual-point contact; by setting heuristic F/T, the alignment of the edge, plane and slots of EC is driven, i.e. the alignment of the plug and socket. Next, we analyse the geometric constraint states at these stages and formulate the corresponding desired contact F/T strategies. To our knowledge, this may be the first attempt to achieve the robotic assembly of ECs under large initial deviations solely using F/T information. Finally, experiments on a UR5 robot indicate that the proposed strategy exhibits robustness to large initial pose deviations and can overcome obstacles caused by friction and jamming to ensure the robotic assembly of ECs. Note to Practitioners—The automatic assembly of ECs in confined spaces with visual occlusions remains an unsolved challenge, especially in aerospace scenarios where manual assembly may not be appropriate. The proposed assembly strategy is based on F/T perception without visual assistance, which can effectively address the perception and assembly processes of plugs under large initial pose deviations, demonstrating robust performance. With the requirement of only a F/T sensor and no need for precise dimensions of the assembly object, this strategy exhibits favourable deployment characteristics for automatic assembly platforms. Furthermore, the proposed strategy can be optimized by the integration of learning-based methods into the perception process, thereby further improving assembly efficiency.