正在加载视频...

视频加载失败

Navigation has been stuck in the same loop for 20+ years: Perception → Localization → Mapping → Planning → Control Every box adds latency. Every box adds failure. LoGoPlanner breaks the loop

10,891 次观看 • 10 个月前 •via X (Twitter)

7 条评论

Robots Digest 🤖 的头像
Robots Digest 🤖10 个月前

The key insight: don’t estimate state—learn geometry. LoGoPlanner finetunes a video geometry backbone to predict metric-scale 3D structure, letting localization emerge implicitly while building dense scene memory for obstacle-aware planning.

Robots Digest 🤖 的头像
Robots Digest 🤖10 个月前

Result: a fully end-to-end diffusion planner that outperforms oracle-localization baselines by +27.3%, generalizes across robots (Go2, G1, TurtleBot), and works in the real world.

Robots Digest 🤖 的头像
Robots Digest 🤖10 个月前

paper: site:

Chuck Petras 的头像
Chuck Petras10 个月前

@BrianRoemmele @GrandpaRoy2

NICHIMEI_PRECISION 的头像
NICHIMEI_PRECISION10 个月前

👍

shibam (🇮🇳/acc) 的头像
shibam (🇮🇳/acc)9 个月前

@Kushu____

United Records 的头像
United Records10 个月前

interesting

相关视频