Current Result Document :
ÇѱÛÁ¦¸ñ(Korean Title) |
·Îº¿ ¹°µû¸£±â Á¶ÀÛ ½ºÅ³ ÇнÀÀ» À§ÇÑ ½Ã¹Ä·¹ÀÌ¼Ç ±â¹Ý ÈÆ·Ã ÇÁ·¹ÀÓ¿öÅ© ±¸Çö |
¿µ¹®Á¦¸ñ(English Title) |
Implementation of Simulation based Training Framework for Learning Robotic Pouring Water Skill |
ÀúÀÚ(Author) |
±èÅ¿ì
Àå¹Î¼ö
±èÀçÈ«
Taewoo Kim
Minsu Jang
Jaehong Kim
|
¿ø¹®¼ö·Ïó(Citation) |
VOL 45 NO. 01 PP. 2565 ~ 2567 (2022. 06) |
Çѱ۳»¿ë (Korean Abstract) |
|
¿µ¹®³»¿ë (English Abstract) |
Imitation learning is one of the preferred approaches for learning complex and long-horizon robotic tasks, however, gathering the dataset for it requires lots of time and cost. To reduce such a cost, in this paper, we introduce our implementation result of training framework which is cost efficient in dataset collection. Through the GPU-accelerated simulator Isaac Gym and our object-oriented path generation method, we could achieve that gathering 50GB demonstration dataset within 25 minutes, in a randomized environment. We validated the dataset generated from our framework by robotic pouring water skill learning. |
Å°¿öµå(Keyword) |
|
ÆÄÀÏ÷ºÎ |
PDF ´Ù¿î·Îµå
|