Research On Task Strategy Of Flexible Assembly Based On Human-robot Cooperation

The change speed of product is fast under the small batch and customized production mode.However,the traditional manual assembly method is inefficient and cannot meet the rapidly changing requirements of personalized customization.At the same time,the assembly of complex products is time-consuming and laborious.The human-robot cooperative assembly(HRCA)mode fully takes the respective advantages of human and robot,which has become one of the key methods for flexible and intelligent assembly.Aiming at the main challenges of HRCA,this paper studies the modelling method of HRCA environment and formulates the cooperation strategy to improve the flexibility of assembly operations.The main research work includes:(1)Aiming at the multi-scale and dynamicity of HRCA environment,a digital twin environment modeling method for HRCA is proposed.Firstly,the depth camera is used to collect the environmental RGB-D information and the three-dimensional assembly environment is recognized based on the improved octree-based point cloud segmentation method.With the designed XML structured file,the element information of point cloud space can be stored after segmentation.Then the Coppeliasim simulation platform is adopted to construct the virtual unit of digital twin assembly,which realizes the interactive control of spatial data towards physical and virtual assembly space.(2)Aiming at the difficulties in designing HRCA strategy for complex products which is over-relying on manual experience,a planning method for HRCA strategy is proposed.Firstly,a point cloud feature recognition model based on the encoder-decoder structure is constructed to obtain the assembly features of assembly parts.Then an assembly task encoding method including feature,relationship and connection mode is also proposed.The task type corresponding to the assembly subtask is determined and a suitable candidate set of HRC sequence is obtained.Finally,the hierarchical reinforcement learning(RL)model is built to realize the dual planning of HRC action sequence and the robotic motion path.(3)Aiming at the lack of flexibility in assembly line for small batch customization,a strategy transferring method of HRCA is proposed.Firstly,the feature distribution distance of dataset between source domain and target domain is calculated.Then the reinforcement learning model or domain adaptation network is built to dynamically adjust the parameters of feature encoder in target domain.The similarity measurement of task attribute involving assembly feature,constraint relations and assembly mode is proposed,which can reuse the existing sequence of assembly job.Finally,the paper carries out case verification through the HRCA work with different types of power lithium batteries.The corresponding experiments are designed and the results show that the action sequence planned by the proposed method is efficient and safe.The assembly strategy transferring method effectively shortens the design time of assembly strategy in target domain and improves the flexibility of HRCA.