Sign In

Talk2DM: Enabling Natural Language Querying and Commonsense Reasoning for Vehicle-Road-Cloud Integrated Dynamic Maps with Large Language Models

Created by
  • Haebom
Category
Empty

์ €์ž

Lu Tao, Jinxuan Luo, Yousuke Watanabe, Zhengshu Zhou, Yuhuan Lu, Shen Ying, Pan Zhang, Fei Zhao, Hiroaki Takada

๐Ÿ’ก ๊ฐœ์š”

๋ณธ ์—ฐ๊ตฌ๋Š” ์ฐจ๋Ÿ‰-๋„๋กœ-ํด๋ผ์šฐ๋“œ(VRC) ํ˜‘๋ ฅ ์ž์œจ์ฃผํ–‰ ์‹œ์Šคํ…œ์˜ ํ•ต์‹ฌ ์ธํ”„๋ผ์ธ ๋™์  ์ง€๋„(DM)์— ๋Œ€ํ•œ ์ž์—ฐ์–ด ์ธํ„ฐํŽ˜์ด์Šค ๋ถ€์žฌ ๋ฌธ์ œ๋ฅผ ํ•ด๊ฒฐํ•˜๊ณ ์ž ํ•ฉ๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด VRCsim ์‹œ๋ฎฌ๋ ˆ์ด์…˜ ํ”„๋ ˆ์ž„์›Œํฌ์™€ VRC-QA ๋ฐ์ดํ„ฐ์…‹์„ ๊ตฌ์ถ•ํ•˜๊ณ , LLM์˜ ์ƒ์‹ ์ถ”๋ก  ๋Šฅ๋ ฅ๊ณผ ์ธ๊ฐ„ ์ •์˜ ๊ทœ์น™์„ ๊ฒฐํ•ฉํ•œ Talk2DM ๋ชจ๋“ˆ์„ ์ œ์•ˆํ•ฉ๋‹ˆ๋‹ค. Talk2DM์€ ๋†’์€ ์ž์—ฐ์–ด ์งˆ์˜ ์ •ํ™•๋„๋ฅผ ์œ ์ง€ํ•˜๋ฉฐ ๋‹ค์–‘ํ•œ LLM๊ณผ์˜ ํ†ตํ•ฉ์ด ๊ฐ€๋Šฅํ•จ์„ ์‹คํ—˜์ ์œผ๋กœ ์ž…์ฆํ–ˆ์Šต๋‹ˆ๋‹ค.

๐Ÿ”‘ ์‹œ์‚ฌ์  ๋ฐ ํ•œ๊ณ„

โ€ข
VRCsim ๋ฐ VRC-QA ๋ฐ์ดํ„ฐ์…‹ ๊ตฌ์ถ•์„ ํ†ตํ•ด ๋™์  ์ง€๋„ ๊ธฐ๋ฐ˜ ์งˆ์˜์‘๋‹ต ๋ฐ ์ถ”๋ก  ์—ฐ๊ตฌ์˜ ๊ธฐ๋ฐ˜์„ ๋งˆ๋ จํ–ˆ์Šต๋‹ˆ๋‹ค.
โ€ข
์ œ์•ˆ๋œ Talk2DM ๋ชจ๋“ˆ์€ LLM์˜ ์ƒ์‹ ์ถ”๋ก  ๋Šฅ๋ ฅ๊ณผ ๊ทœ์น™ ๊ธฐ๋ฐ˜ ์ ‘๊ทผ ๋ฐฉ์‹์„ ๊ฒฐํ•ฉํ•˜์—ฌ ์ž์—ฐ์–ด ์งˆ์˜ ์ดํ•ด๋„๋ฅผ ๋†’์˜€์Šต๋‹ˆ๋‹ค.
โ€ข
๋‹ค์–‘ํ•œ LLM๊ณผ์˜ ์œ ์—ฐํ•œ ํ†ตํ•ฉ ๋ฐ 2-5์ดˆ์˜ ๋น ๋ฅธ ์‘๋‹ต ์‹œ๊ฐ„์œผ๋กœ ์‹ค์šฉ์ ์ธ ์ ์šฉ ๊ฐ€๋Šฅ์„ฑ์„ ๋ณด์—ฌ์ฃผ์—ˆ์Šต๋‹ˆ๋‹ค.
โ€ข
๋” ํฐ ๋ชจ๋ธ์ด ์ •ํ™•๋„๋ฅผ ๋†’์ด์ง€๋งŒ ํšจ์œจ์„ฑ์ด ์ €ํ•˜๋˜๋Š” trade-off ๊ด€๊ณ„๊ฐ€ ์กด์žฌํ•˜๋ฉฐ, ์ด๋Š” ํ–ฅํ›„ ๋ชจ๋ธ ๊ฒฝ๋Ÿ‰ํ™” ๋ฐ ์ตœ์ ํ™” ์—ฐ๊ตฌ์˜ ๊ณผ์ œ๊ฐ€ ๋  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
๐Ÿ‘
Made with Slashpage