Recently, the acceptance results of IROS2020 (IEEE/RSJ International Conference on Intelligent Robots and Systems 2020) were announced. Two scientific research achievements of Wu Yanmin and Wang Peng, graduate students of 2018 from the Faculty of Robot Science and Engineering of NEU have been accepted.
The paper Eao-slam: Monocular Semi-Dense Object SLAM Based on Ensemble Data Association was jointly completed by NEU, the Chinese University of Hong Kong (CUHK) and Ulster University of the UK, with Wu Yanmin as the first author, professor Zhang Yunzhou as the corresponding author, and NEU as the first completion unit. The paper, with robot environment perception as the research object, first puts forward a kind of integrated data association strategy in response to the different measurements in SLAM, and then uses the isolated forests and line segment alignment algorithm to estimate the object pose. It also proposes a method of lightweight semantic map construction and finally creates a semi-dense semantic map and an object-oriented lightweight map. This research has broken through the limitations of semantic SLAM and is of great significance to realizing accurate decision-making and interaction in the process of robot perception, navigation and capture.
The paper TP-TIO: A Robust Thermal-Inertial Odometry with Deep Thermal Point was jointly completed by the REALLAB Research Group from NEU’s Faculty of Robot Science and Engineering and the Robotics Institute of Carnegie Mellon University, with Wang Peng as the second author, NEU as the second completion Unit, and associate professor Fang Zheng and professor Sebastian Scherer as the academic advisors. The paper puts forward a new feature point extraction network on infrared images - Thermal Point. While improving the real-time performance, it can maintain precision. In addition, in a new way of network training, it applies simulated and real fixed pattern noise to the training sets to improve the network's ability to resist fixed pattern noise of infrared images. Finally, it is applied to the Visual Inertial Odometry (VIO). This method makes it possible for robots to achieve robust pose estimation in extreme scenarios such as smoke.
IROS, co-organized by the Institute of Electrical and Electronics Engineers (IEEE) and the Robotics Society of Japan (RSJ), is one of the three top international conferences in the field of robotics. With the theme of “Consumer Robots and Our Future” this year, it collects researches from such related fields as robotics and AI, robot vision, sensors and cloud robots. IROS has been run for about three decades, and the achievements exhibited at it have laid a solid foundation for the sustainable development of robot technologies, products and services. It is known that the REALLAB Research Group has published articles at IROS for two consecutive years. Last year, the paper A Robust Laser-Inertial Odometry and Mapping Method for Large-Scale Highway Environments was collected by IROS2019, of which the first author is Zhao Shibo, a master graduate of 2019 from NEU’s Faculty of Robot Science and Engineering, and the corresponding author is associate professor Fang Zheng. The study was jointly completed by NEU, Tencent Unmanned Driving Laboratory and Carnegie Mellon University, with NEU as the first completion unit. In this paper, a novel laser inertial odometer and mapping method is proposed, which can achieve real-time low drift and robust attitude estimation in large-scale highway environments. This research has solved the problem of pose estimation of unmanned vehicles in high-speed and dynamic scenarios, and advanced the application of SLAM theory.