PILOT: A Perceptive Integrated Low-level Controller for Loco-manipulation over Unstructured Scenes

์ €์ž: Xinru Cui, Linxi Feng, Yixuan Zhou, Haoqi Han, Zhe Liu, Hesheng Wang | ๋‚ ์งœ: 2026-01-24 | DOI: 10.48550/arXiv.2601.17440 📄 PDF


Essence

Figure 1

Fig. 1. Method overview of PILOT. We propose a unified single-stage reinforcement learning framework that seamlessly int

PILOT๋Š” humanoid robot์˜ loco-manipulation์„ ์œ„ํ•œ ํ†ตํ•ฉ ๋‹จ๊ณ„ RL ํ”„๋ ˆ์ž„์›Œํฌ๋กœ, ์ง€๊ฐ ๊ธฐ๋ฐ˜ locomotion๊ณผ ์ „์‹  ์ œ์–ด๋ฅผ ๋‹จ์ผ policy๋กœ ํ†ตํ•ฉํ•˜์—ฌ ๋น„์ •ํ˜• ์ง€ํ˜•์—์„œ ์•ˆ์ •์ ์ธ ์ž‘์—… ์‹คํ–‰์„ ๊ฐ€๋Šฅํ•˜๊ฒŒ ํ•œ๋‹ค.

Motivation

Achievement

Figure 3

Fig. 3. Real-world Experiments. PILOT successfully executes object transport tasks across challenging terrains. The robo

How

Figure 2

Fig. 2. Visualization of expert activation across six motion modes. The

Originality

Limitation & Further Study

Evaluation

Novelty: 4/5 Technical Soundness: 3/5 Significance: 4/5 Clarity: 4/5 Overall: 4/5

์ดํ‰: PILOT๋Š” humanoid loco-manipulation ๋ฌธ์ œ์— ๋Œ€ํ•œ ํ†ตํ•ฉ์ ์ด๊ณ  ์‹ค์šฉ์ ์ธ ํ•ด๊ฒฐ์ฑ…์„ ์ œ์‹œํ•˜๋ฉฐ, cross-modal perception๊ณผ MoE ๊ตฌ์กฐ๋ฅผ ํ†ตํ•ด ๊ธฐ์ˆ ์  ๊ธฐ์—ฌ์™€ ์‹ค์ œ ๋กœ๋ด‡ ๊ตฌํ˜„์˜ ์„ฑ๊ณต์  ์‚ฌ๋ก€๋ฅผ ๋ณด์—ฌ์ค€๋‹ค.

๊ฐ™์ด ๋ณด๋ฉด ์ข‹์€ ๋…ผ๋ฌธ

๋‹ค๋ฅธ ์ ‘๊ทผ
์ €์ˆ˜์ค€-๊ณ ์ˆ˜์ค€ ํ†ตํ•ฉ ์ œ์–ด์™€ ๋‹ฌ๋ฆฌ Perceptive Integrated Low-level Controller๋ฅผ ํ†ตํ•ด ์ „์‹  ๋ณดํ–‰ ์ œ์–ด๋ฅผ ์ง์ ‘ ํ•™์Šตํ•˜๋Š” ๋ฐฉ์‹์„ ๋ณด์—ฌ์ค€๋‹ค.
← ๋ชฉ๋ก์œผ๋กœ ๋Œ์•„๊ฐ€๊ธฐ

๐ŸŽง Audio Overview

์ด ๋…ผ๋ฌธ ๋ฆฌ๋ทฐ๋ฅผ ํŒŸ์บ์ŠคํŠธํ˜• ์˜ค๋””์˜ค๋กœ ์ƒ์„ฑํ•ฉ๋‹ˆ๋‹ค. (Gemini ยท ํ‚ค๋Š” ๋ธŒ๋ผ์šฐ์ €์—๋งŒ ์ €์žฅ ยท ์™„์„ฑ๋ณธ์€ ์ด๋ฉ”์ผ๋กœ๋„ ์ „์†ก)
โ–ธ ๊ณ ๊ธ‰: ๊ตฌ์„ฑ ๋ฐฉํ–ฅ(๋Œ€๋ณธ ์ž‘์„ฑ ์ง€์นจ) ์ง์ ‘ ์ˆ˜์ •