Shitij Govil retweeted
Excited to share FloMo! We explore how predicting 3D motion instead of video can better help robots learn directly from human video.
Really enjoyed working on this with @jerthesquare_ @NamraByTheSea @shitijgovil @animesh_garg
Check out our work on representing “motion” as 3D scene flow with a video generation backbone. XYZ->RGB trick for using pretrained WAN VAE might be my favorite idea 😆
Learned so so much on this project, great collaborators and mentors: @jerthesquare_ @NamraByTheSea @agarwal_ayush05 @animesh_garg
Shitij Govil retweeted
Most world action models learn to act by predicting videos.
We find that predicting how the scene moves in 3D performs better.
Introducing FloMo: a world action model that uses 3D motion as the bridge between video pre-training and robot control.
Shitij Govil retweeted
Introducing MESA: how well do robot policies generalize to new layouts, objects, and task compositions?
Benchmark four axes of generalization on single-arm and bimanual robots with only ~5-7K training demos each, plus a pipeline to generate your own tasks and data. 🦾
Shitij Govil retweeted
Reducing diffusion steps can surprisingly improve WAM performance.
More steps add action jitter, but single-step denoising yields smooth actions efficiently (12 Hz, 9GB VRAM).
Here's how from our CoRL 2026 paper TAPESTRY, done during my internship last year @ToyotaResearch 🧵👇
Shitij Govil retweeted
Can novel-view synthesis (NVS) be a strong pretext task for learning geometry-aware representations?
We introduce SNAP, a multi-view representation learning framework that performs NVS in the latent spaces of pre-trained image teachers instead of RGB.
1/n
snap-nvs.github.io
We’re excited to welcome @animesh_garg as our new Chief Research Officer at @microagi, with the team from @usablerobots joining him.
He has spent most of his career working on one of the hardest problems in robotics: getting robots out of the lab and into the real world.
Frontier AI now works. Hardware is becoming cheap. And the industry is finally learning how to collect robot data at scale.
What’s still missing is post-training.
Taking a general-purpose model and teaching it one specific job, in one specific environment, until a company can actually depend on it.
In LLMs, post-training is what turned impressive demos into products. In robotics, that layer barely exists yet.
A strong model is a capability. A robot that can perform the same job, on the same line, every shift, without an engineer standing next to it, is a product.
And that process has to happen again for every task, every site, and every customer.
We believe the company that can do this well, fast, and at scale will be the company that puts robots into production at massive scale.
MicroAGI is building both sides of that system.
With @joinshift, we’re building one of the most diverse and largest egocentric datasets in the world.
With Atlas, we post-train and harness models directly on enterprise deployments.
Our goal is simple:
Deploy 1 million robots.
Shitij Govil retweeted
You fine-tune a robot foundation model on a hard task. It gets 25%. Now what?
Introducing Q-Planning, a learning-based harness that lets large black-box robot policies recursively self-improve.
On a hard fine-grained task: 25% → 80% in 100 robot attempts (~30 mins), no extra human data.
q-planning.github.io/
1/n 🧵
Shitij Govil retweeted
🤖 Scaling deployment in robotics can be difficult and expensive. How can we properly serve a VLA/WAM to 10+ robots concurrently from a single GPU?
Introducing Armory, a system for serving large robot policies to a fleet of robots from the cloud/edge. Through real-time scheduling of action chunks, Armory boosts task throughput over baselines by up to 18% on heterogeneous fleets in the real world.
🧵1/6
Shitij Govil retweeted
Robotics is currently bottlenecked by the lack of large-scale datasets, compared to the billions of images and trillions of text tokens fueling foundation models in computer vision and NLP. Why? Collecting robot demonstrations is prohibitively time-consuming and expensive.
Introducing COBALT: A cloud-based teleoperation platform designed to democratize robot learning at scale.
Shitij Govil retweeted
The code is finally open source!
github.com/pairlab/AMPLIFY
Remember this thread about teaching robots from videos without action labels?
→ Use human videos to teach robots
→ 27x better task generalization
→ Easily plug in your own datasets
Time to build 🚀
Robotics data is expensive and slow to collect.
Robotics labs and companies spend months just to collect around 10k hours of demonstration data, all while that much video is uploaded to YouTube every 20 minutes. However, none of this video data contains action labels. How can we bridge the gap?
AMPLIFY solves this problem by learning Actionless Motion Priors that unlock better sample efficiency, generalization, and scaling for robot learning.









