Saurabh Nair retweetledi
The Eternal Sloptember geohot.github.io//blog/jekyll/u…
English
Saurabh Nair
399 posts
Saurabh Nair
@lazy_marcus
AI / Robotics. Applied Scientist @wayve_ai. Previously @vicariousai, @CMU_robotics
San Francisco, CA Katılım Ekim 2011
681 Takip Edilen237 Takipçiler
Saurabh Nair retweetledi
Saurabh Nair retweetledi
Today, we’re announcing a strategic technology partnership with @Stellantis, bringing the Wayve AI Driver to the STLA AutoDrive platform.
The collaboration will enable hands-free, door-to-door supervised automated driving across both urban streets and highways.
➡️ Launching in 2028 in North America.
📈 Built on an AI architecture that can scale across vehicle platforms and markets over time.
Learn more 👇
wayve.ai/press/stellant…
English
Saurabh Nair retweetledi
WARNING: Robotics is not for the faint of heart. There are no silver bullets here. It is tiresome, unglamorous, and at times brutal work. The only way out is through, and the way through is unknown. But on the other side awaits insatiable demand and unimaginable opportunity.
English
Saurabh Nair retweetledi
2/4 of our LA-Pose Showcase. 🌧️
Today - a highway exit and roundabout scenario with shifting road geometry and stormy conditions. As the vehicle approaches and navigates the turn, LA-Pose remains stable and consistent, while VGGT struggles and visibly deforms while navigating the roundabout.
Where does this gap in performance come from? After understanding what motion looks like, we freeze the inverse dynamics model and train a lightweight pose estimation head on top which learns to translate those latent motion representations into precise 3D estimates: translation, rotation, field-of-view, and metric scale in a single forward pass.
This unique approach helps LA-Pose achieve over 10% higher pose accuracy than state-of-the-art feed-forward methods, using orders of magnitude less labeled data.
Read through our full paper: la-pose.github.io
English
Saurabh Nair retweetledi
Meet LA-Pose. Our latest model taking Wayve another step towards generalization at scale.
LA-Pose employs large-scale self-supervised learning, building strong motion representations for 3D perception from 10.2 million unlabeled driving video snippets, unlike today's strongest approaches that often depend on expensive, carefully curated 3D supervision.
With only a lightweight pose head and limited labelled data, LA-Pose achieves:
📷 State-of-the-art camera pose estimation
🌎 Strong zero-shot generalization across diverse driving scenarios
🏷️ Orders of magnitude less labelled data than fully supervised 3D approaches
Our full blog post: wayve.ai/thinking/la-po…
Explore the full paper here: la-pose.github.io
English
Saurabh Nair retweetledi
Day 1/4 of our LA-Pose Showcase. 📷
To show what our latest model, LA-Pose, is capable of, we're sharing examples of how it anticipates motion and trajectory across diverse driving scenarios, comparing it to the state-of-the-art supervised pose estimation method, VGGT.
Today - a night-time highway driving scene on long curved roads and repeated headlight glare. While LA-Pose keeps the curve coherent, VGGT consistently drifts and loses the route shape.
How? LA-Pose uses an inverse dynamics model, trained on 10.2 million unlabelled driving clips, to learn what we call a latent action: a compact representation of how the world shifted between consecutive frames. The model is never told the vehicle’s speed or heading. It learns to infer them in a self-supervised way.
Stay tuned for more.
English
The result? A method that beats traditional supervised-at-scale models while being orders of magnitude more data-efficient. 📉✨
English
Excited to share our #CVPR 2026 paper: LA-Pose! 🚀
We’re introducing a "freeze & train" Camera-Pose-Estimation baseline by leveraging Self-Supervised World-Modeling.
Wayve@wayve_ai
Meet LA-Pose. Our latest model taking Wayve another step towards generalization at scale. LA-Pose employs large-scale self-supervised learning, building strong motion representations for 3D perception from 10.2 million unlabeled driving video snippets, unlike today's strongest approaches that often depend on expensive, carefully curated 3D supervision. With only a lightweight pose head and limited labelled data, LA-Pose achieves: 📷 State-of-the-art camera pose estimation 🌎 Strong zero-shot generalization across diverse driving scenarios 🏷️ Orders of magnitude less labelled data than fully supervised 3D approaches Our full blog post: wayve.ai/thinking/la-po… Explore the full paper here: la-pose.github.io
English
Saurabh Nair retweetledi
LA-Pose: Latent Action Pretraining Meets Pose Estimation
Zhengqing Wang, Saurabh Nair, Prajwal Chidananda, Pujith Kachana, Samuel Li, Matthew Brown, @YasutakaFuruka1
tl;dr: learn a latent action space without pose labels, then reuse those motion-centric features for camera pose estimation
arxiv.org/abs/2604.27448
English
Thinking of hosting more chai and samosas in SF at ours ☕️
Want to come hang out with good people and have fun snacks, lmk below! 🙌
English
Checkout our #neurips2025 spotlight paper. We enhance capabilities of 3D foundation models like Dust3R with camera-metadata that can be acquired from a robot.
Wayve@wayve_ai
🔷 Introducing Rig3R - our new geometric foundation model for 3D perception in AVs. wayve.ai/thinking/rig3r/
English
Saurabh Nair retweetledi
Saurabh Nair retweetledi
Meta releases DINOv3
Everyone talks about Llama but I think Meta's contributions to computer vision (SAM, DINOv2, etc.) are highly underappreciated.
They're now releasing a newer iteration with large model (7B param), better data curation, and improved dense features.
This is sure to be the foundation for many computer vision use-cases going forward.
AI at Meta@AIatMeta
Introducing DINOv3: a state-of-the-art computer vision model trained with self-supervised learning (SSL) that produces powerful, high-resolution image features. For the first time, a single frozen vision backbone outperforms specialized solutions on multiple long-standing dense prediction tasks. Learn more about DINOv3 here: ai.meta.com/blog/dinov3-se…
English
Are you in the Bay Area tomorrow (specifically willing to go to Berkeley) like a good mystery, are down to clown and available from 11:30am-2pm?
Let me know. I’m hosting a mystical, magical, musical Monday.
English
Saurabh Nair retweetledi
The world is in desperate need of more beautiful environments.
Current cities are like dementors for the soul.
One of my favorite companies, @Monumental_Labs, is leading the change.
Had a call with them today about making a marble bust and estimating the cost of building a temple.
The future is about having a vision, generating it in 3D, and getting robots to build it.
Micah Springut@mspringut
There’s beauty here and I feel sorry for those who can’t see it.
English
Saurabh Nair retweetledi
Saurabh Nair retweetledi
Niels Bohr's original drawing of the structure of the Radium atom [1922]
English
Saurabh Nair retweetledi
Saurabh Nair retweetledi
NEWS: Morgan Stanley analyst Adam Jonas said he was left "speechless" after he spent an hour this week being driven in a robotaxi in London.
Adam rode in a development vehicle from Wayve that used 7 cameras, no radar, no LIDAR, no HD maps and no labeling. It was all end-to-end AI-enabled foundation model.
He said it drove with confidence and like an experienced London cab driver.
Recent video from Wayve of a 75 minute uninterrupted autonomous drive in central London:
English