Á¤º¸°úÇÐȸ ÄÄÇ»ÆÃÀÇ ½ÇÁ¦ ³í¹®Áö (KIISE Transactions on Computing Practices)
ÇѱÛÁ¦¸ñ(Korean Title) |
ÀÚÀ²ÁÖÇàÀ» À§ÇÑ ¸ÖƼ¿¡ÀÌÀüÆ® ½ÉÈ °ÈÇнÀ |
¿µ¹®Á¦¸ñ(English Title) |
Multi-agent Deep Reinforcement Learning for Autonomous Driving |
ÀúÀÚ(Author) |
ÀÌÈ«¼®
¹ÚÀº¼ö
±è½ÂÀÏ
Hongsuk Yi
Eunsoo Park
Seungil Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 24 NO. 12 PP. 0670 ~ 0674 (2018. 12) |
Çѱ۳»¿ë (Korean Abstract) |
ÀÚÀ²ÁÖÇàÀº µµ·Î¿¡¼ Â÷¼± º¯°æ, Ãß¿ù, ¾çº¸ µîÀ» ÇÒ ¶§ Á¤±³ÇÑ »óȲÆÇ´Ü ±â¼úÀ» Àû¿ëÇØ¾ß ÇÏ´Â ¸ÖƼ-¿¡ÀÌÀüÆ® ¹®Á¦ÀÌ´Ù. ÀÚÀ²ÁÖÇàÂ÷·®µéÀÇ ¿¬¼ÓÀûÀÎ ÇൿÀ» Á¦¾îÇϱâ À§ÇÏ¿© º» ³í¹®¿¡¼´Â ½ÉÈ °áÁ¤·ÐÀû Á¤Ã¥ °æ»ç °ÈÇнÀ ¾Ë°í¸®ÁòÀ» Àû¿ëÇÏ¿´´Ù. À̸¦ À§ÇÏ¿© Â÷¼±º¯°æÀÌ ºó¹øÈ÷ ¹ß»ýÇÏ´Â µµ·Î ȯ°æÀ» ½Ã¹Ä·¹ÀÌÅÍ·Î ±¸ÇöÇÏ¿´°í, °ÈÇнÀ¿¡¼ Àû¿ëµÈ º¸»óÀº °³º° Â÷·®ÀÌ ¸ñÀûÁö Â÷¼±¿¡ µµÂøÇÏ¸é ³ôÀº º¸»óÀ» ¹ÞÁö¸¸, Â÷·®ÀÌ ´Ù¸¥ ¸ñÀûÁö Â÷¼±¿¡ µµÂøÇÒ °æ¿ì³ª Â÷·®³¢¸® Ãæµ¹ÀÌ ¹ß»ýÇÒ °æ¿ì¿¡´Â ¹úÄ¢À» ¹Þµµ·Ï ¼³°èÇÏ¿´´Ù. 16°³ÀÇ ¸ÖƼ-¿¡ÀÌÀüÆ® Â÷·®À» ÇнÀÇÑ °á°ú ÇнÀ½Ã°£ÀÌ ÃæºÐÇÒ¼ö·Ï Â÷¼±º¯°æÀ» Á¦¾îÇÒ ¼ö ÀÖÀ½À» ¾Ë¾Ò´Ù. ÇÏÁö¸¸ ½ÉÈ °ÈÇнÀ°ú ½Ã¹Ä·¹ÀÌÅÍ È¯°æÀÇ ÇÑ°è·Î, ÇнÀÀÌ ÁøÇàµÇ´Â °úÁ¤¿¡¼ º¸»ó °ªÀÌ ±Þ°ÝÈ÷ °¨¼ÒÇÏ¿´°í, ÀÌ·Î ÀÎÇÏ¿© Â÷·® ÁÖÇàÀº ¸Å¿ì ºÒ¾ÈÁ¤ÇÑ ÁÖÇàÀ» ÇÏ¿´´Ù.
|
¿µ¹®³»¿ë (English Abstract) |
Autonomous driving is a multi-agent problem, wherein the host vehicle must adopt sophisticated human driving negotiation skills with other drivers on the road when overtaking, giving away. In this paper we apply deep reinforcement learning to the problem of forming long-term driving. More specifically, we use deep deterministic policy gradient algorithms, termed actor- critic algorithm. A reward function promoting longitudinal velocity, while penalizing transverse velocity and divergence from the track center, is used to train multi-agents. The actor-critic algorithm was trained and evaluated in a synthetic environment. Results reveal that our deep reinforcement learning approach can generalize and adapt well to weaving sections on real roads.
|
Å°¿öµå(Keyword) |
ÀΰøÁö´É
ÀÚÀ²ÁÖÇà
½ÉÈ °ÈÇнÀ
¸ÖƼ-¿¡ÀÌÀüÆ®
Artificial intelligence
autonomous driving
deep reinforcement
multi-agent
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|