Matteo Merler

Artificial Intelligence Researcher @ FBK NLP, previously MSc @ Aalto University

[Matteo Merler @merler_m]

Hungry for more Astra robot videos? 🤖

How about a large-scale sandboxed evaluation to go with it? We ran 98,000 evaluations across 28 simulated environments and found some surprisingly clever physical reasoning along the way!

New preprint 🧵👇 (1/9)

[Matteo Merler @merler_m]

Hungry for more Astra robot videos? 🤖

How about a large-scale sandboxed evaluation to go with it? We ran 98,000 evaluations across 28 simulated environments and found some surprisingly clever physical reasoning along the way!

New preprint 🧵👇 (1/9)

[Matteo Merler @merler_m]

@chrisdotai Thank you Opus 5.5 for the nice infographics, it will probably be the next one that has to go through our environments😁

[Matteo Merler @merler_m]

Hungry for more Astra robot videos? 🤖

How about a large-scale sandboxed evaluation to go with it? We ran 98,000 evaluations across 28 simulated environments and found some surprisingly clever physical reasoning along the way!

New preprint 🧵👇 (1/9)

[Matteo Merler @merler_m]

@chrisdotai It's also worth noting that agents still have interactive access to the simulator, so they can use that to generate variants and search for edge-cases. For example, in the paper we show how in this environment agents explicitly look for hard seeds and use them for testing.

[Matteo Merler @merler_m]

@rb_rupam We mainly used the environments from the KinDER benchmark (https://t.co/i3W5zH3qw3), they are mostly implemented in Mujoco/PyBullet, so no generative models involved on that side

[Matteo Merler @merler_m]

This is my first paper with @tomssilver's group, and I'm super happy with the results. Thank you @Bw_Li1024 @thosehippos @yichao_liang @WangRobin58334 @YixuanHuang13 for all the help! 🙏

Excited to keep working together! 🚀

🔗 Paper + videos: https://t.co/znk0mOmh9f (9/9)

[Matteo Merler @merler_m]

Over the last six months, coding agents have changed how I do research. This project has made me realize how much we need to understand them, beyond using them as tools.

It's also helped me understand my own research taste and find a direction I'm really excited about. (8/9)🧵

[Tom Silver @tomssilver]

@basisorg and my group at Princeton are recruiting a postdoc to work at the intersection of robotics and code-based world models.

We’re looking for someone excited about abstractions & planning, and who knows their way around a real robot.

Link 👇 Thanks for boosting!

[Tom Silver @tomssilver]

New blog post:

https://t.co/T29LEH7lpY

This is the second in a "series" of posts about how to define and work on research problems. (The first post was 3 years ago...)

This one addresses the agentic elephant in the room.

[Yixuan Huang @YixuanHuang13]

🤖 Curious about the connection between Task and Motion Planning (TAMP) and robot learning?

📄 Check out our new survey on Learning By and For Task and Motion Planning!

We organize the growing literature on TAMP + learning, introduce a taxonomy of TAMP learning, clarify the state of the art, and highlight promising directions for future research. 🚀

🔗 https://t.co/uqha9QP4gf

[Sergio Hernández @SergioHzGz]

Long-horizon tasks + sparse rewards = no supervision. So, we need dense supervision methods… but how do we evaluate them?

Meet QVal, our latest release👇

[Yixuan Huang @YixuanHuang13]

Meet KinDER — a stress test for robot physical reasoning. All 13 methods failed 😈

🌎 25 environments
♾️ Infinite tasks
🏋️ Gymnasium API
⚒️ Over 20 parameterized skills
🪧 Human demonstrations
📊 13 baselines (planning and learning)

From @Princeton @CMU_Robotics @ICatGT @CambridgeMLG @nvidia @MIT_CSAIL

🧵 1/n

[Minttu Alakuijala @malakuijala]

I'm happy to announce our paper
Generating Code World Models with Large Language Models Guided by Monte Carlo Tree Search, has been accepted @NeurIPSConf'24!

Link: https://t.co/R8Si9oohje
Joint work with @dainese_nicola, @MerloMerler and @marttinen_pekka.

#NeurIPS2024