S-Agent 2.0 gives coding agents spatial investigation tools, lifting ReVSI by 20.5 points
jiqizhixin · x · 2026-11-09
An arXiv preprint from NTU, HKU and Ropedia tackles why video assistants confuse left/right and distance despite fluent descriptions. S-Agent 2.0 equips a general coding agent with perception tools to build a persistent, queryable 3D scene state — no model training needed. With Qwen3.8-27B held fixed, ReVSI rises from 61.8 to 82.3 (+20.5 points), turning spatial questions into explicit code operations rather than verbal intuition.
More from coding & agent
- Day 309/365 of GPU Programming: Optimizing LLaMA2-7B on Edge with SIMD and Multithreading — NandoDF · 2026-11-09
- Berkeley-led paper: evolved agents gain 5.2 reliability points while losing 10 tasks they once solved — rohanpaul_ai · 2026-11-09
- Running a Home Mac mini as an Always-On Node for Remote AI Tasks — xiaohu · 2026-11-09
- Capwise repo goes public: native macOS screen recording and AI editing workbench — vista8 · 2026-11-09
- Capwise open-sources beta of AI screen recorder with bilingual captions and Agent CLI — vista8 · 2026-11-09
- Agent PRs merge at just 32.7% vs 84.4% for humans, with 91% more review time — gerardsans · 2026-11-09