An LSTM-based Framework for Motion Pattern Recognition of Handling Robots

• Published in: 2022 7th International Conference on Intelligent Computing and Signal Processing (ICSP) pp. 1184 - 1187
• Main Authors: Li, Fangyu, Zheng, Jianbin, Yan, Zhiqi, Tang, Yong, Chen, Liufei, Wang, Zhuorui, Gao, Yifan, Wang, Yu
• Format: Conference Proceeding
• Language: English
• Published: IEEE 15.04.2022
• Subjects: Correlation; Costs; handling robots; long short-term memory network; Machine learning; motion pattern recognition; Neural networks; Pattern recognition; Service robots; Signal processing
• DOI: 10.1109/ICSP54964.2022.9778425

With the development of the construction industry, construction work more and more tends to mechanization, automation, and intelligence. The appearance of handling robots greatly improves the handling efficiency and reduces the cost. Even though great efforts have been made on motion pattern recognition for handling robots, the existing methods still have some shortcomings. Previous studies are all used for motion pattern recognition, each input is processed separately, which cuts off the time continuity between the latter data and the former data. To this end, in this paper, we first emphasize the importance of temporal correlation between data in the field of motion pattern recognition and devise an LSTM based method to ease the issue above. Finally, we use Softmax to effectively calculate the classification results of data potential features and temporal information. Compared with traditional machine learning models and popular neural network models, extensive experiments on real-world datasets demonstrate the effect of our framework is superior to the baselines, thus providing novel perspectives to the building automation industry.