Dimensional

Augmented Reality is the natural interface for robotics. Not only for giving commands, but also to visualise hardware data (like LiDAR) natively in 3D. @specs AR glasses, @UnitreeRobotics Go2 Lite quadruped, @dimensionalos OS for the robot

Spots are limited. luma.com/dn72ypdy

Hosting an open house at Dimensional SF office! If you ever wanted to operate an actual robot, now's the chance, join us to: - Drive robot dogs, humanoids and arms - Meet Dimensional team and robotics enthusiasts - Take on fun challenges to win prizes Register below⬇️

@AnInternetter 👾

@dimensionalos real life video game

Big launch coming next week! As a teaser, watch our teammates in India, China, and Argentina use dimOS to remotely command a robot dog through SF's 4th of July fireworks.

@swstica 🐶

@dimensionalos let's go🚀🦾

Tomorrow @stash_pomichter will give sneak peek into our hosted teleoperation on discord Community Hour, 9am PST. Join now⬇️ discord.gg/dimos

Full implementation⬇️ github.com/dimensionalOS/…

Our @UnitreeRobotics G1 now knows what its arms can reach before it ever moves We sampled 2 million poses per arm and built a reachability map and the planning setup went from ~100 ms to ~62 microseconds. 4.9% false positives, verified against IK Fully open source

Join the event here discord.gg/dimos

Community hour on discord🧠🤖 Every Friday 9am PST/6pm CET 📆 This week, @stash_pomichter will talk about the latest release on navigation. Join in, chat, bring your questions. It's gonna be fun, we promise❤️ Whether you're building in robotics, looking to move or just curious, there's something for all.

Announcing the first production robot navigation framework on $500 hardware Explore the world once → your robot agent will relocalize and build a persistant, spatial memory across sessions SLAM, relocalization, loop closure, map i/o, planning, control No ROS. Open source.

5/ Pack Mind won the Navigation track by building an operational memory for a robot pack. Simultaneous Localization and Mapping (SLAM) is the go-to technique for a mobile robot to build awareness of an unknown space. It works well for one robot, but gets brittle across a swarm. Merging separate SLAM maps requires recognizing when two robots have seen the same place, then using that overlap to align their maps. A single bad match corrupts the shared map. Merging maps across a swarm (collaborative SLAM) is still an active research problem. Pack Mind coordinates through a shared semantic ledger of zone names and findings, so there's no map to merge and no coordinate frame to align. Because that memory lives in the ledger and not in any one dog, the pack survives failure: a dog drops, its discoveries persist, and a teammate inherits its zones. Congratulations to @sharqwy, and the builders behind Pack Mind 🏆

4/ Fetch is an autonomous brand ambassador: a robot dog that hands out cold Cokes and instant printed photos. A vision model reads the camera on every frame and decides where to move, what to say, and when to take the shot. The team benchmarked vision and speech models to keep the latency around one second. The photo is captured from the dog's camera, composited, and printed into the guest's hands on the spot. Fetch shows how robot agents will deliver enterprise utility in the future, becoming the AI age equivalent of vending machines. Congratulations to @seifip, @Wenjix and @GuoZhuoRan, the builders behind Fetch 🏆

How fast can software engineers go from bits to atoms? We ran the 1st in-person dimOS hackathon to find out 15 teams had 48h and 10 Unitree Go2s to deploy code into the physical world. These are our favorite submissions 🧵

Every week new robots integrated by the community!

2/ Shoutout to the team @PimVandenBosch2 and contributors @kevin_zakka for work on mjlab / mink frameworks used here. And as always, fully open source: github.com/dimensionalOS/…

Claude Fable 5 integrated reachability analysis into our G1 Manipulation framework Now the humanoid knows what its arms can reach before it ever tries Features: 5D G1 arm capability map in a gravity-aligned built from 2M FK samples per arm with self-collision checking Planning context creation dropped from 100ms to 62µs 14 joint dual-arm control, solving QP differential IK for both arms