具身智能真机数据集建设的现实图景、发展掣肘与化解策略OACHSSCD

[Purpose/Significance]As one of the key directions of future industrial cultivation,embodied intelligence has become a strategic field to achieve new economic growth and seize new heights of development.In the process of pro-moting the industrialization of embodied intelligence,real-world dataset plays a pivotal role.The research on the construc-tion of embodied intelligence real-world dataset is of great significance for releasing the multiplier effect of the embodied intelligence real-world dataset and enabling the intelligent butterfly change of thousands of industries.[Methods/Process]Based on the comprehensive application of literature research,case analysis,expert consultation and other methods,on the basis of explaining the concept connotation of the embodied intelligence real-world dataset,the overall construction pattern and construction details of the embodied intelligence real-world dataset construction are described from the macro level by means of multi-case display.At the same time,the AGIBOT WORLD dataset is selected for single case analysis,and its construction mechanism is revealed from a micro perspective.Further,it focused on the practical constraints of the current development of the embodied intelligence real-world dataset,and proposed a resolution strategy.[Result/Conclusion]At present,the construction of the embodied intelligence real-world dataset mainly has the development pain points of"one high,one low and three risks",that is,the data cost is high,the data reuse rate is low,and the high-quality real-world dataset circulation faces the triple risks of public safety,intellectual property rights and personal privacy.Based on this,the resolution strategies proposed in this paper mainly include:using the mixed acquisition mode of real-world data and simulation data,using ontology-free acquisition technology,accelerating the construction of dataset standards,implementing the technical scheme of"standardized data conversion middleware+residual learning fine-tuning",introducing hierarchical and classified supervision system,constructing rights and interests protection network,and improving the design of privacy computing technology.