AUTHOR=Han Zhuqin 

TITLE=Multimodal intelligent logistics robot combining 3D CNN, LSTM, and visual SLAM for path planning and control

JOURNAL=Frontiers in Neurorobotics

VOLUME=Volume 17 - 2023

YEAR=2023

URL=https://www.frontiersin.org/journals/neurorobotics/articles/10.3389/fnbot.2023.1285673

DOI=10.3389/fnbot.2023.1285673

ISSN=1662-5218

ABSTRACT=In today's ever-growing logistics landscape, intelligent robots play a key role in improving efficiency, reducing costs and enhancing safety. Traditional path planning methods are difficult to adapt to changing environments, which may lead to problems such as collisions and conflicts. This research aims to solve the problem of path planning and control of logistics robots in complex environments. The method proposed in this study can integrate information from different perception modalities to achieve more accurate path planning and obstacle avoidance control, and improve the autonomy and reliability of logistics robots. The method proposed in this paper first uses 3D CNN to learn the feature representation of objects in the environment, so as to realize the object recognition function. Then, the spatio-temporal features are modeled by LSTM to predict the behavior and trajectory of dynamic obstacles. This helps the robot to more accurately predict the future position of obstacles in complex environments, thereby avoiding potential collision risks. Finally, the Dijkstra algorithm is used for path planning and control decisions to ensure that the robot chooses the optimal path in different scenarios. In a series of experiments, the method proposed in this paper performs well in both path planning accuracy and obstacle avoidance performance, achieving significant improvements over traditional methods. This intelligent path planning and control scheme will effectively improve the practicability of logistics robots in complex environments, and promote the efficiency and safety of the logistics industry.