• Àüü
  • ÀüÀÚ/Àü±â
  • Åë½Å
  • ÄÄÇ»ÅÍ
´Ý±â

»çÀÌÆ®¸Ê

Loading..

Please wait....

±¹³» ³í¹®Áö

Ȩ Ȩ > ¿¬±¸¹®Çå > ±¹³» ³í¹®Áö > Çѱ¹Á¤º¸°úÇÐȸ ³í¹®Áö > Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)

Á¤º¸°úÇÐȸ³í¹®Áö (Journal of KIISE)

Current Result Document :

ÇѱÛÁ¦¸ñ(Korean Title) Ä¿¸®Å§·³ ·¯´×À» ÀÌ¿ëÇÑ Ç×°ø±â °­È­ÇнÀ
¿µ¹®Á¦¸ñ(English Title) Aircraft Reinforcement Learning using Curriculum Learning
ÀúÀÚ(Author) ¹èÁ¤È£   °­À±¼º   À±¼®¹Î   ±è¿ë´ö   ±è¼ºÈ£   Jung Ho Bae   Yun-Seong Kang   Sukmin Yoon   Yong-Duk Kim   Sungho Kim  
¿ø¹®¼ö·Ïó(Citation) VOL 48 NO. 06 PP. 0707 ~ 0712 (2021. 06)
Çѱ۳»¿ë
(Korean Abstract)
Àü ¼¼°èÀûÀ¸·Î ¹«ÀÎ Ç×°ø±â Á¦¾î¸¦ À§ÇÑ ´Ù¾çÇÑ ¾Ë°í¸®ÁòÀÌ Á¦¾ÈµÇ¾úÁö¸¸ Ž»ö½Ã°£, ºÒ¸íÈ®ÇÑ ±ÔÄ¢ µîÀÇ Á¦ÇÑ»çÇ×ÀÌ ÀÖ¾ú´Ù. À̸¦ ±Øº¹Çϱâ À§ÇÏ¿© º» ¿¬±¸¿¡¼­´Â ½ÉÃþ °­È­ÇнÀÀ» ÀÌ¿ëÇÑ Á¦¾î ±â¹ýÀ» Àû¿ëÇÏ°í È¿°úÀûÀÎ ÇнÀÀ» À§ÇÏ¿© Á¶ÁØ°¢µµ(ATA) ±â¹ÝÀÇ Ä¿¸®Å§·³ ·¯´×À» Á¦¾ÈÇÑ´Ù. Á¦¾ÈÇÑ ±â¹ýÀÇ È¿¿ë¼ºÀ» È®ÀÎÇϱâ À§ÇÏ¿© Ç×°ø±â ÁúÁ¡ 6-ÀÚÀ¯µµ ¸ðµ¨À» Àû¿ëÇÑ 3Â÷¿ø ½Ã¹Ä·¹ÀÌ¼Ç È¯°æÀ» ±¸ÃàÇÏ°í µÎ Ç×°ø±â°¡ ¼­·Î µÚµ¹¾Æ º¸°í ÀÖ´Â »óȲ¿¡¼­ °ÝÃßÇÏ´Â ½Ã³ª¸®¿À¸¦ ÇнÀÇÏ¿´´Ù. ±× °á°ú Ä¿¸®Å§·³ ·¯´×À» Àû¿ëÇÏÁö ¾Ê¾ÒÀ» °æ¿ì ATA 60¡Æ ÀÌ»óÀº Á¦ÇÑµÈ ½Ã°£¿¡ ÇнÀÀÌ ¿Ï·áµÇÁö ¾ÊÀº °ÍÀÌ ¹ÝÇØ ATA ±â¹Ý Ä¿¸®Å§·³ ·¯´×À» Àû¿ëÇÑ °æ¿ì¿¡´Â º°µµÀÇ º¸»óÇÔ¼ö Ãß°¡ ¾øÀÌ 180¡Æ±îÁö ÇнÀÀÌ ¿Ï·áµÇ´Â °ÍÀ» È®ÀÎÇÏ¿´´Ù.
¿µ¹®³»¿ë
(English Abstract)
Diverse algorithms have been proposed to control unmanned aircrafts. However, they have limitations such as long exploration time and/or unclear behavior rules. To overcome the drawbacks and for efficient training, we propose an aircraft control technique using deep reinforcement learning applying antenna train angle (ATA) based curriculum learning. To validate the effectiveness of the proposed technique, we constructed a 3D simulation environment adapting a 6-DOF aircraft point model and conducted training with an initial setting of two fighters in the neutral position situation where they are looking back. The results showed that the proposed technique can achieve the goal of ATA 180¡Æ when the fighters are looking back without adding supplemental reward functions, while the deep reinforcement learning (DRL) without ATA curriculum could not succeed the learning over ATA 60¡Æ in a limited training time.
Å°¿öµå(Keyword) ½ÉÃþ°­È­ÇнÀ   Ä¿¸®Å§·³·¯´×   Ç×°ø±â Á¦¾î   Á¶ÁØ°¢µµ   deep reinforcement learning   curriculum learning   aircraft control   ATA  
ÆÄÀÏ÷ºÎ PDF ´Ù¿î·Îµå