Loading...

Çмú´ëȸ

Korea Robotics Society

  >   Çмú´ëȸ   >   ±¹³»Çмú´ëȸ

±¹³»Çмú´ëȸ

±âÁ¶°­¿¬
[±âÁ¶°­¿¬I]
ÁÖÁ¦
Rethinking VLAs: Bridging the Manipulation Gap
ÀϽÃ
2026³â 2¿ù 5ÀÏ(¸ñ) 11:20~12:05
Àå¼Ò
2Ãþ ±×·£µåº¼·ë Æ÷·¹½ºÆ®È¦
°­¿¬ÀÚ
¹ÚÁ¾¿ì(¼­¿ï´ëÇб³/±³¼ö)
¿ä¾à
·Îº¿ Á¶ÀÛÀº ´Ù¾çÇÑ Çü»ó°ú ¹°¼ºÀ» Áö´Ñ ¹°Ã¼¸¦ ´ë»óÀ¸·Î Æø³ÐÀº ÆÄÁö (prehensile) ¹× ºñÆÄÁö(nonprehensile) ÀÛ¾÷À» ¼öÇàÇØ¾ß ÇÑ´Ù. À̸¦ À§ÇØ Á¢ÃË »óÅ¿¡¼­ÀÇ ¿îµ¿°ú ÈûÀ» µ¿½Ã¿¡ Á¤¹ÐÇÏ°Ô Á¶À²ÇÏ´Â Á¦¾î°¡ ¿ä±¸µÇ´Âµ¥, ·Îº¿°ú ¹°Ã¼ÀÇ Á¤È®ÇÑ ¸ðµ¨, ±×¸®°í ½Å·ÚÇÒ ¼ö ÀÖ´Â ¼¾½ÌÁ¤º¸°¡ Á¦°øµÈ´Ù¸é ÀüÅëÀûÀÎ ¸ðµ¨ ±â¹Ý ±â¹ýÀ¸·Îµµ ÀÌ·¯ÇÑ ¹®Á¦¸¦ ¿øÄ¢ÀûÀ¸·Î ÇØ°áÇÒ ¼ö ÀÖ´Ù. ±×·¯³ª ½ÇÁ¦ ȯ°æ¿¡¼­´Â ¸¶Âû, º¯Çü, ¿¹±âÄ¡ ¾ÊÀº Á¢ÃË, ¼¾¼­ ÀâÀ½°ú °°Àº ºÒÈ®½Ç¼ºÀÌ Á¸ÀçÇϸç, ÀÌ·Î ÀÎÇØ °úµµÇÑ ¸ðµ¨¸µ ³ë·Â°ú ¼öÀÛ¾÷ Æ©´×ÀÌ ÇÊ¿äÇÒ»Ó´õ·¯, »õ·Î¿î ·Îº¿À̳ª ÀÛ¾÷À¸·ÎÀÇ ÀϹÝÈ­°¡ ¾î·Æ´Ù.
ÀÌ·¯ÇÑ ÇѰè´Â ·Îº¿ Á¶ÀÛÀ» À§ÇÑ foundation model¿¡ ´ëÇÑ °ü½ÉÀ» Ã˹߽ÃÄ×´Ù. ±×·¯³ª ÃÖ±Ù ÁÖ¸ñ¹Þ´Â vision-language-action (VLA) model ±â¹Ý Á¢±Ù¹ýÀº ¿©·¯ Á¦¾àÀ» ¾È°í ÀÖ´Ù. Á¶ÀÛ µ¥ÀÌÅÍ´Â Èñ¼ÒÇϰí Çϵå¿þ¾î ÀÇÁ¸ÀûÀ̸ç, ´ë±Ô¸ð ¼öÁýÀÌ ¾î·Æ´Ù. ¶ÇÇÑ ¹°¸® ½Ã¹Ä·¹ÀÌÅÍ´Â ¸¶Âû°ú º¯ÇüÀ» Æ÷ÇÔÇÑ º¹ÀâÇÑ Á¢ÃË Çö»óÀ» Á¤È®È÷ ¸ðµ¨¸µÇÏ´Â µ¥ ÇѰ谡 ÀÖ´Ù. ´õ ³ª¾Æ°¡, ´ëºÎºÐÀÇ VLA ¸ðµ¨Àº À§Ä¡ Áß½ÉÀÇ Ãâ·Â ±¸Á¶¿¡ ÀÇÁ¸ÇÏ¿© Á¶ÀÛÀÇ º»ÁúÀûÀÎ Á¢ÃË ±â¹Ý Ư¼ºÀ» ÃæºÐÈ÷ ¹Ý¿µÇÏÁö ¸øÇÑ´Ù.
È¿°úÀûÀÎ ·Îº¿ Á¶ÀÛ foundation modelÀ» ±¸ÃàÇϱâ À§Çؼ­´Â ±Ùº»ÀûÀ¸·Î ´Ù¸¥ °èÃþÀû ¾ÆÅ°ÅØÃ³°¡ ÇÊ¿äÇÏ´Ù.  ÀÌ ¾ÆÅ°ÅØÃ³´Â ·Îº¿ °èȹ ¹× Á¦¾î, ¿ªÇаú ¹°¸®, ±×¸®°í human motor control¿¡¼­ À¯·ÊÇÑ ¿©·¯ inductive biasµéÀ» Æ÷ÇÔÇϸç, ƯÈ÷ Èû°ú º¯À§¸¦ ÅëÇÕÀûÀ¸·Î Á¶ÀýÇÔÀ¸·Î½á compliantÇÑ µ¿ÀÛÀ» °¡´ÉÇÏ°Ô ÇÑ´Ù.  º» °­¿¬¿¡¼­´Â Á¦¾ÈÇÏ´Â »õ·Î¿î ¾ÆÅ°ÅØÃ³ÀÇ ±¸Á¶¿Í ±¸¼º¿ä¼­¸¦ »ìÆìº¸¸ç, Á¶ÀÛ foundation modelÀ» ±¸ÃàÇϱâ À§ÇÑ °úÁ¤°ú ÁÖ¿ä µµÀüÀ» °íÂûÇØ º»´Ù.
¾à·Â
- 1995~ ÇöÀç ¼­¿ï´ëÇб³ ±â°è°øÇкΠ±³¼ö
- 2022~2023 IEEE Robotics and Automation Society ȸÀå
- 2013~2018 IEEE Transactions on Robotics ÆíÁýÀå
- 2025~2028 IEEE Board of Directors (Division X director-elect/director)
- 1991~1995 University of California Irvine ±â°èÇ×°ø°øÇкΠÁ¶±³¼ö
- 2018~ ÇöÀ砢߼¼ÀÌÁö Founder & CEO
ÇзÂ
- 1985-1991 Harvard U. ÀÀ¿ë¼öÇÐ ¹Ú»ç
- 1981~1985 MIT Àü±âÄÄÇ»ÅͰøÇÐ Çлç
 
 
[±âÁ¶°­¿¬2]
ÁÖÁ¦
Advancing AI Agents towards Solving Real-World Problems
ÀϽÃ
2026³â 2¿ù 6ÀÏ(±Ý) 11:00~11:45
Àå¼Ò
2Ãþ ±×·£µåº¼·ë Æ÷·¹½ºÆ®È¦
°­¿¬ÀÚ
ÀÌÈ«¶ô(LG AI¿¬±¸¿ø °øµ¿¿¬±¸¿øÀå/¹Ì½Ã°£´ëÇб³ ±³¼ö)
¿ä¾à
AI technology is evolving beyond generating information into the era of "Agentic AI," where systems autonomously plan and act within digital environments. In this keynote, I will discuss the development journey of LG AI Research's flagship foundation model, EXAONE, and introduce the technical foundation of the recently unveiled K-EXAONE. Building on this, I will trace the logical progression of intelligent agents from digital workflows to the physical world. Specifically, I will start with enterprise applications like ChatEXAONE and move toward advanced Web and Computer Use Agents, highlighting research efforts to overcome challenges in visual grounding, reasoning, and long-horizon planning. I will also address how these advancements in agentic AI can serve as R&D partners in expert domains such as chemistry and life sciences to enhance research efficiency. Finally, I will briefly touch upon "Physical AI," discussing how the capabilities of large-scale models can be integrated with foundational paradigms of autonomous behavior to enable more versatile and robust robotic systems.
¾à·Â
- 2025³â~ÇöÀç LG AI¿¬±¸¿ø °øµ¿¿¬±¸¿øÀå
- 2020³â~ÇöÀç LG AI¿¬±¸¿ø Chief Scientist of AI (CSAI), ºÎ»çÀå
- 2023³â~ÇöÀ砹̱¹ ¹Ì½Ã°£´ëÇб³ ÄÄÇ»ÅͰøÇаú Á¤±³¼ö (Full Professor)
- 2016³â~2020³â ±¸±Û ºê·¹ÀÎ(Google Brain) Research Scientist
- 2016³â~2023³â ¹Ì±¹ ¹Ì½Ã°£´ëÇб³ ÄÄÇ»ÅͰøÇаú ºÎ±³¼ö (Associate Professor)
- 2010³â~2016³â ¹Ì±¹ ¹Ì½Ã°£´ëÇб³ ÄÄÇ»ÅͰøÇаú Á¶±³¼ö (Assistant Professor)
ÇзÂ
- 2010 ¹Ì±¹ ½ºÅÄÆÛµå ´ëÇб³ ÄÄÇ»ÅͰúÇÐ ¹Ú»ç (Ph.D. in Computer Science)
- 2006 ¹Ì±¹ ½ºÅÄÆÛµå ´ëÇб³ ÄÄÇ»ÅͰúÇÐ ¹× ÀÀ¿ë¹°¸®ÇÐ ¼®»ç (M.S. in CS and Applied Physics)
- 2003 ¼­¿ï´ëÇб³ ¹°¸®ÇÐ ¹× ÄÄÇ»ÅͰúÇÐ Çлç (º¹¼öÀü°ø)