Inside a real-time 3D mapping system! 🧭 That's how modern home bots map and localize using only cameras. @maticrobots is using voxel-based neural networks running on NVIDIA Jetson Orin to build real-time, photorealistic 3D maps of the world around its robots. Its autonomy

See, maps are not going away. Not everything needs to be end to end. But using a map with your vla is great for a multi purpose robot

https://t.co/aQYPNC1ZPr

Scaling Image Geo-Localization to Continent Level Philipp Lindenberger @pesarlin Jan Hosang, Matteo Balice @mapo1 Simon Lynen, Eduard Trulls tl;dr: combine ground image with aerial to get cell-prototype. Acc similar to ~1.5Tb ground photo retrieval @ 42Gb https://t.co/61CSw7r5s6

From Lamaria paper on SLAM: online calibration is crucial. Even an ideally perfect SLAM system without online calibration would fail as the sequence gets longer

Some LiDAR mounted vehicles from Hangzhou. I didn’t expect mapping vehicles to enter such places that are generally challenging to visit, but they are expanding the boundaries.

Introducing Generative View Stitching (GVS), a non-autoregressive sampling method for length extrapolation of video diffusion models. GVS enables collision-free camera-guided video generation for predefined trajectories, including Oscar Reutersvärd's Impossible Staircase (1/9).

AMEN

“Payable”

🔥Join us tomorrow, Friday, at 4:15 pm CET, for a talk by Prof. Chelsea Finn @chelseabfinn @Stanford on Developing Generalist Robots at ETH Distinguished Seminar in Robotics, Systems, and Control! 📽️Zoom: https://t.co/MGc4XLmUv1 🔗Schedule & Recordings: https://t.co/24rBOZ8yuA

Dear algorithm, Please show this tweet to the coolest founders on X, roboticists, researchers, nerds, and creators. I want to follow a bunch.

Struggling with low shelf worth? AI copilots like Cactus are proven to boost performance and shelf esteem. Try Cactus today! 🌵

@_ashwanirathee For the video stream, we currently use the VGGT lifted point cloud and compress it by grid sampling. That is the visual result you see from our demo video. Regarding the SLAM point cloud, we tried in our Sonata paper; it also works, but we didn't do this at scale. Maybe we could

Introducing Concerto 🎶 Joint 2D-3D Self-Supervised Learning Emerges Spatial Representations. What is it: Concerto is a self-supervised Point Transformer V3 that jointly learns from 2D and 3D modalities, producing rich spatial representations. It can take both point clouds and

Reasoning models have massively expanded what LLMs are capable of, but this hasn’t necessarily applied to robotics. Perhaps this is in part because robots need to reason over space, not just words and symbols; so the robotics version of a reasoning model would need to think in

Just make it ≈4.5k$ and ship it fast

Whatever it is I got my budget ready

✨️🚀Japan Mobility Show🚗✨️ #モビショー 2025 HONDAのロケット展示を撮影してきました〜！ 大樹町で垂直離着陸実験をしていたあのロケット実機！ なんとここまで寄れます❣️ グリッドフィンもばっちり、脚も機体もじっくり見られるし絶対見に行って！！ 一般公開日は10/30〜11/9だよ✨️

The Real2Sim is so good that it takes some time to tell what is real and what is sim.

@KhurramJaved_96 @fahadjibla We've worked a lot with @graphcoreai which is the right kind of chip (>1000 tiles, each with around 1MB local SRAM and all-to-all communication), and very programmable, though also data-centre targeted. Tenstorrent also have very interesting chips, more edge oriented @changh95