👉 Pointing is our first “language”—babies master it before words. Precise spatial grounding powers robotics, assistive tech, HCI, and vision-language interfaces. 🤔 But can today's MLLMs point with pixel-level accuracy and truly ground visual reasoning?📷We introduce PointArena,

Great to see Molmo leading on pointing👉

Let us know how good is Molmo is at language guided pointing 👈 Vote here👇

Excited to be at #CVPR2025 in Nashville! 🎉 I’m presenting a demo paper with real-world robot demos and co-organizing two workshops: Robo 3D VLM and Generalization for Robotic Manipulation. Let’s connect if you’re into 🔥 Robotics + AI — and don’t miss our stacked speaker

Building on our work with Molmo, we’re excited to introduce GraspMolmo — a vision-language model that predicts semantically meaningful grasps conditioned on natural language. A fantastic effort led by our PYI, Abhay Deshpande !

Following up on our work on Molmo: Molmo points, but how can those points power real-world robotics? Introducing GraspMolmo, VLM that plugs seamlessly into robotic systems to generate semantically meaningful grasp poses from natural language commands. 👉 abhaybd.github.io/GraspMolmo/

I am doing something silly by testing whether I can remember and deliver multiple talks on the same day on different slices of my group’s research. If you are at #CVPR2025 on June 11th, come to one or all of them :D 9:05am: Behaviors & bodies: how they shape one another

Our Molmo work won Best Paper Honorable mention at #CVPR2025 ! This large project was one of my best experiences with a fantastic team!

🥳 Excited to share that I’ll be joining the CS Department at UNC-Chapel Hill (UNC Computer Science UNC AI) as an Assistant Professor starting Fall 2026! Before that, I’ll be working at Ai2 Prior (Ai2 Prior @ AI2) and UW (Allen School) on multimodal understanding and generation.

🚨Tired of binary pass/fail metrics that miss the bigger picture? 🤖Introducing #RoboEval — an open benchmark that shows *how* robot manipulation policies behave and *why* they fail, not just *if* they succeed. 🧵1/n 🔗 robo-eval.github.io 📄 robo-eval.github.io/media/RoboEval…

It’s incredible to have both your advisors at the same company! With Dieter Fox building the Robotics team, and Ranjay Krishna leading PRIOR, Ai2 is set to become a powerhouse in robotics, computer vision, and embodied AI for open science research . Excited to be part

Excited to present our work at #ICML next week! Always happy to chat about all things 🔥 in Robotics and AI. I’m also be on the academic job market this coming year — would love to connect about any potential opportunities!

We’re presenting SAM2Act at #ICML! Come check out the many amazing projects from AI2, and stop by to chat with us and learn more about our work.

🤖✨ What if models that take action in the physical world could think through your instructions? Meet MolmoAct, our new fully open Action Reasoning Model (ARM) that does just that. 🧵

Reasoning is central to purposeful action. Today we introduce MolmoAct — a fully open Action Reasoning Model (ARM) for robotics. Grounded in large-scale pre-training with action reasoning data, every predicted action is interpretable and user-steerable via visual trace. We are

We are launching MolmoAct🤖✨ A fully open Action Reasoning Model (ARM) that can reason in space: it perceives → it plans → it acts. 🧵👇

This to me really feels like how robot foundation models "should" work. i like that it can autoregressively predict depth tokens, lift to 2.5d, and use this for reasoning - it feels like a true robotics analogue of modern reasoning LLMs. Really exciting work.

First fully open Action Reasoning Model (ARM); can ‘think’ in 3D & turn your instructions into real-world actions: [📍 Bookmark for later] A model that reasons in space, time, and motion. It breaks down your command into three steps: ✅ Grounds the scene with depth-aware

Most AI models still think in words. People, without even noticing, think with our bodies, planning how to move, grasp, and use things around us. MolmoAct brings that to robotics: reasoning in space before acting. This is how we will get to the GPT-moment for robotics.

We’re humbled by the overwhelming interest in MolmoAct! 🚀 At AI2, we’re committed to full open-source — data, training, evaluation, real-world robot system setup code, and more. We have a super-detailed technical report covering what worked (and what didn’t). Soon, you’ll have