> $# \pJava Excel API v2.6 Ba==h\:#8X@"1Arial1Arial1Arial1Arial + ) , * `V"DC,title[*]contributor[author]contributor[advisor]keywords[*]date[issued] publisher citationsidentifier[uri]identifier[doi]abstractrelation[journal]relation[volume]relation[no]relation[page]A study on the Position Estimation of the 1.3m tall Bipedal Humanoid Robot in a Noisy environment through Keypoint-based LocalizationJunYoung Kim\լǌ2022. 8\ՑYPxhttp://hanyang.dcollection.net/common/orgView/200000626586;
https://repository.hanyang.ac.kr/handle/20.500.11754/174591;This paper proposes an estimation of a 1.3m tall humanoid robot's position in a noisy environment using only a camera sensor.
The robot's behavior decision is an essential element of the autonomous robot in the future. Moreover, to satisfy that element, the position estimation of the robot is a fundamental prerequisite. Localization (position estimation) determines the relative position of the robot within the map environment. Existing mobile robots have been extensively studied in localization. However, while mobile robots use sensors that are capable of omnidirectional sensing like LiDAR, humanoids have limited constraints. They can only use cameras with narrow viewing angles, making the localization problem more challenging. Also, localization is challenging because humanoids do not move as stable as mobile robots, and camera view frames oscillate from side to side. Developing localization with high accuracy under the above-limited constraints is necessary.
In this paper, several algorithms were applied to improve the localization performance. First, a keypoint-based Monte Carlo localization using YOLO was developed, which is a deep learning-based object detection model. In addition, the data association(matching) process was improved from the nearest neighborhood matching algorithm to the Hungarian algorithm, which is widely used in autonomous driving. Through this, data association performance was improved, and also was successful in estimating the robot's position in real-time. Next, an algorithm called Augmented Monte-Carlo localization was developed to solve the robot kidnap problem, which frequently occurs in noisy environments. It improved the algorithm to be able to more robustly estimate the robot's position.
The developed algorithm was applied to the biped humanoid robot ALICE3 in the HERoEHS lab to verify the localization performance. A soccer field was installed in both the Webots Simulation and Real World to verify whether the robot could estimate its position in real-time. It is expected that the developed algorithm will positively influence the behavior decision of the humanoid robot ALICE3 in RoboCup 2022.| |8 tT| <̹D tǩX xtǈ\ Xֽ 1.3mX 48xtܴ \t X X| ` ǔ LବD H\.
\X ٳ(behavior decision)@ ^<\ ̹ (ȉ \X ɔ\ ƌt. ƌ| q¤0 t \X X @ Iǈ ɔ\ ptt. \|ttX(X )@ Xֽ H \X X| XՔ t. 0t | \@ \|ttX ( ι@ l ɉ T. X̹ | \@ |t@ t ) \ <| tǩXՔp t 48xtܴ \x |D tT|̹D tǩt| \䲔 \ ptt Ǵ \|ttX 8Ȕ T L̰X. \ 48xtܴ | \ Ʉt Ht J Ȱ\ ٳX0 L8 X t L̰XՌ . @ @ \ pt @ Uĳ| \|ttX t DՔX.
|8@ \|ttXX 1D ĳTX0 t LବD ȩX. < %ݲ 0X ̀ xx YOLO| X xǸ 0 Lt|\ \|ttXD X. \ pt0 iD 0t greedy\ Lବ<\ L$8 ǔ nearest neighborhood m ) (ȉ ιt ǔ H Lବ<\ X i D p. t| t <\ t Xĳ] ̹@
!"Root EntryWorkbook&SummaryInformation(DocumentSummaryInformation8