Mahbod Tajdini

CV Researcher @microagi | prev. @ETH, @EPFL

[Mahbod Tajdini @Ace3Zi]

@supertoday @toolshed_labs @notgiannei Thanks

[Mahbod Tajdini @Ace3Zi]

@supertoday @toolshed_labs @notgiannei We are now iterating to improve what we have

[Mahbod Tajdini @Ace3Zi]

@supertoday @toolshed_labs @notgiannei https://t.co/Wt1q276wWM

[Mahbod Tajdini @Ace3Zi]

@BigZaddyKan3 @rohamzn 🫶🏼🫶🏼🫶🏼

[Mahbod Tajdini @Ace3Zi]

@supertoday @toolshed_labs @notgiannei Not yet!

[Lukas Ziegler @lukas_m_ziegler]

A head-mounted camera is now robot training data! 📹

@microagi announced MicroSLAM, built by their computer vision team in Zürich. 

It's monocular SLAM from human egocentric video, and they report first place on the LaMaria benchmark for monocular SLAM, and second against any SLAM system built so far, including ones running many more cameras and IMUs.

Monocular egocentric video is the cheapest robotics data that exists. One person, one camera, walking a factory floor. 

MicroSLAM turns that footage into interactive 3D scenes, RL environments that behavioural foundation models can be trained inside, so a robot learns to navigate a real facility from video a human recorded.

microagi says the existing SLAM systems broke for them in three specific ways:

→ The static world assumption falls apart in dynamic egocentric video, since almost everything around the wearer is in motion
→ Scale ambiguity, because a single camera can't tell you how big anything actually is
→ Long-horizon temporal consistency, and productive robotics sequences run from tens of minutes to hours

ORB-SLAM3, the usual baseline, hits all three.

Their two fixes are dynamic-object-aware recognition that identifies what's moving so the static world assumption survives, and a loop closure system built on rich features from the environment.

Congrats @ZenoInMotion, @bercankilic, @YoanIlievX, @notgiannei and the whole team! 😮‍💨

~~

♻️ Join the weekly robotics newsletter, and never miss any news → https://t.co/GoA3ZuwoPB

[Mahbod Tajdini @Ace3Zi]

@whiskeysourxr 👀

[Mahbod Tajdini @Ace3Zi]

@bercankilic 🚀🚀🚀

[Bercan @bercankilic]

A great first step toward making large-scale mobile manipulation deployments possible; well done to our Zurich lab for the breakthrough!

The best non-hardware-bound SLAM on earth to our knowledge.

[Mahbod Tajdini @Ace3Zi]

@makeshifted @JulienBlanchon @notgiannei It was supported by @microagi from a long time ago!

[Mahbod Tajdini @Ace3Zi]

@worksalt @gazorp5 @notgiannei 88 minutes?

More like 8months ;)

[Mahbod Tajdini @Ace3Zi]

@toolshed_labs @notgiannei Hopefully you will figure it out in the paper 👀

[Mahbod Tajdini @Ace3Zi]

@makeshifted @JulienBlanchon @notgiannei Maybe it will be open sourced 👀

[Gianni Hodel @notgiannei]

Today we're announcing microSLAM, a monocular SLAM system built by our computer vision team in Zürich with ETH Zurich's Computer Vision and Geometry Lab. It ranks first on the LaMaria benchmark for monocular SLAM, and holds up against systems that carry many more cameras and IMUs. Ours runs on a single RGB stream.

Robot learning is bottlenecked on data, and the largest untapped source is people going about their work. A camera on someone's head for an afternoon in a plant is a record of how that plant actually operates, including all the parts nobody writes down.

That data is only useful if you can recover the geometry, and geometry from one moving camera in a world that will not sit still is the hard version of the problem. Solving it monocular is what makes the data cheap. Rigs do not scale to every worker. Glasses do.

microSLAM turns that footage into interactive 3D environments where behavioral foundation models can be trained. One person walks a factory floor. A robot learns to navigate it.

[Mahbod Tajdini @Ace3Zi]

@brunoeducsant @microagi By keeping its poses and fuse a dense depth source through TSDF