Arjun Guru

cs phd @uw, @allen_ai, prev engineering @nvidia, robotics + cs + economics @penn @grasplab

[Jiafei Duan✈️CoRL2026 @DJiafei]

The best model we tested on MolmoSpace-Commonsense succeeds just 9.3% of the time.

Robotics needs benchmarks that go beyond task success to test commonsense reasoning at every step just like how we humans do things.

Introducing MolmoSpace-Commonsense, led by @arjungru! 🤖

We’re nowhere near saturation. More results for GPT-6, Opus, and other frontier models are coming soon!

[Arjun Guru @arjungru]

@prodarhan Yeah honestly that might have been what cooked me

[Arjun Guru @arjungru]

Robot policies look great on today’s benchmarks. But hide the mug behind a few objects, or ask for the hot pan, and they fall apart.

Introducing MolmoSpaces-CommonSense: a benchmark for commonsense reasoning in manipulation.

No policy we tested tops 30% on any task 🧵

[Arjun Guru @arjungru]

@prodarhan Top 10 craziest experiences of my life ngl

[Arjun Guru @arjungru]

(11/n) Finally, a huge thanks to my coauthors and advisors @max_argus, @jjaesungpark, @DJiafei, @RanjayKrishna, and Ali Farhadi!

🌐 Leaderboard (CommonSense tab): https://t.co/00n1zX1xy9
💻 Code: https://t.co/LD667Vhf2y
📄 Paper: https://t.co/5DanhuMtws

[Arjun Guru @arjungru]

(10/n) Takeaway: different architectures fail in the same ways, pointing to shared training data, not models. High scores on clean benchmarks likely overstate real-world readiness.

[Arjun Guru @arjungru]

Can generalist policies handle tasks that need reasoning and not just motor skills? Presenting MolmoSpaces-CommonSense at the SemRob workshop poster session at RSS from 11-11:30 today (room CB11.B3.101). Come say hi!

[Tanush @tanushyy]

Remember action recognition? The days of trying to climb on Kinetics?👻

Announcing VideoNet, a CVPR 2026 Highlight 🎉 which revitalizes action recognition in the VLM era

Explore our data with this fun, interactive demo: https://t.co/W53aBi3QAX

(1/8) 🧵

[Jae Sung Park @jjaesungpark]

VLMs today—including our own Molmo—point via raw text strings (e.g. "<x=0.53, y=0.71>"). What if pointing meant directly selecting the visual tokens instead? 🤔

Introducing MolmoPoint: Better Pointing for VLMs with Grounding Tokens 🎯

🔓models, code, data, demo all OPEN 🧵👇

Paper: https://t.co/vTH5vnLckN

[Maya Guru @mayasguru]

Introducing MolmoBot! We set out with an ambitious goal to make zero-shot sim2real manipulation work. Seeing it transfer to the real world was incredibly exciting. Can’t wait to see what large-scale sim data unlocks next.

[Ai2 @allen_ai]

Introducing MolmoSpaces, a large-scale, fully open platform + benchmark for embodied AI research. 🤖

230k+ indoor scenes, 130k+ object models, &amp; 42M annotated robotic grasps—all in one ecosystem.

[Ethan Shen @ethanlshen]

Today, we release SERA-32B, an approach to coding agents that matches Devstral 2 at just $9,000. It is fully open-source and you can train your own model easily - at 26x the efficiency of using RL.

Paper: https://t.co/aeD6T2WW3O

Here’s how 🧵