Research On Kinematics Parameter Calibration And Non-geometric Error Compensation Of Self-reconfigurable Robot

Self-reconfiguring robots have emerged as a promising solution for enhancing the adaptability and versatility of robots in dynamic and uncertain environments.These robots can change their physical configurations by rearranging their modular units,thereby enabling them to perform a wide range of tasks in different environments.However,the flexibility and variability of the reconfiguration process pose significant challenges in ensuring the end accuracy of self-reconfiguring robots.Specifically,the kinematic model undergoes reshaping after each reconfiguration,which makes it challenging to ensure the end accuracy of the reconfigured end.Furthermore,frequent reconfigurations can lead to interface wear,which can exacerbate the kinematic deviations.In addition,non-geometric error sources,such as robot flexibility deformation and j oint gap,can have a significant impact on the end accuracy of selfreconfiguring robots.The coupling effects of these error sources further complicate the task of ensuring the end accuracy of these robots to meet the desired task requirements.Therefore,the development of effective techniques for mitigating these error sources and ensuring the end accuracy of self-reconfiguring robots remains a crucial research challenge in the field of robotics.Based on the project of BUPT Action Plan to Enhance Capacity for Scientific and Technological Innovation "Research on Dynamic Reconfigurable Homogeneous Modular Robot for In-orbit Service" and the research on kinematic parameter calibration and non-geometric error compensation strategy of self-reconfigurable robot.Specifically,The research includes several components,such as the development of a selfcalibration scheme and a kinematic autonomous modeling method,as well as an investigation into error models,a parameter identification techniques.and a non-geometric error compensation strategy.Details are as follows.Firstly,the sources of error affecting the end accuracy of the selfreconfiguring robot are analyzed and sorted.Subsequently,a selfcalibration scheme is developed for the kinematic parameters of the selfreconfiguring robot based on this analysis.A mathematical representation of the topology of the modular unit and the self-reconfiguring robot is constructed,followed by the establishment of a mapping relationship between the topological description and the kinematic model.The goal is to solve the problem of universal kinematic modeling of the selfreconfiguring robot with variable topology,and provide a model basis for kinematic parameter calibration and non-geometric error compensation.Secondly,the error transfer mechanism of robot kinematic models with varying topologies was meticulously analyzed,followed by the construction of error models for both chain topology robots and tree topology robots,leveraging the local exponential product.The error transfer law of isomorphic topology kinematic parameters was scrutinized to optimize the error model.Addressing the issue of poor identification accuracy and low efficiency of high-dimensional error models,an initial population model of the inter-cell generation algorithm was created.Moreover,a hierarchical identification algorithm,blending L-M algorithm and CMA-ES algorithm,was devised to enhance the accuracy and efficiency of parameter identification.Simulation results demonstrate that after calibration with the error model and hierarchical parameter identification algorithm developed in this research,the accuracy of the end position of the chain topology robot improved by 64.59%,while the attitude accuracy enhanced by more than 87.76%.Similarly,the position accuracy of the tree topology robot improved by 68.32%and 58,72%,respectively,while the attitude accuracy improved by more than 63.64%.Overall,the error model and parameter identification method presented in this research significantly enhance the terminal accuracy of various topological structures.Thirdly,non-geometric errors are typically difficult to model due to their nonlinearity and high level of coupling.This paper proposes using BP neural network,ELM network,and RBFN network to predict and compensate for non-geometric residuals.However,the effectiveness of the ELM network is shown to be heavily dependent on the initial parameter value.To address this issue,an improved ELM network is proposed by integrating it with the PSO algorithm based on VNS algorithm optimization.In addition,Ada_BP,Ada_ELM,and Ada_RBFN networks are constructed using an ensemble learning approach with multiple base learning machines to increase the stability of neural network prediction and compensation.The effectiveness of the six networks is compared using a simulation experiment.The results indicate that the integrated learning network based on the AdaBoost framework significantly improves the precision of terminal pose and ensures a certain level of prediction stability.Finally,to validate the theoretical methods proposed in this research,an experimental platform for the self-reconfiguring robot calibration system was built,and the control software was developed.Physical verification experiments were conducted using the chain topology robot and the tree topology robot built from modular units as the research objects.The experiments were designed to evaluate the effectiveness of the proposed theoretical methods for kinematic parameter calibration and nongeometric error compensation of the self-reconfiguring robot.The results of the physical experiments demonstrate the practicality and effectiveness of the proposed methods.Overall,the research offers a comprehensive approach to improving the end accuracy of self-reconfiguring robots,and the experimental verification of our methods provides strong evidence of their efficacy.