Twitxy Saved

Matteo Merler

@merler_m

PhD Student @ Tübingen AI Center (Bethge Lab), IMPRS-IS & ELLIS. Previously Researcher @ FBK NLP, MSc @ Aalto University
Trento, Trentino-South Tyrol merlerm.github.io Joined August 2016
294 Following 108 Followers 38 Posts

Latest posts 5

Pinned post

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)
0:12
1 replies 22 reposts 120 likes 21K views View on X
Matteo Merler @merler_m
✈️If you are at @corl_conf today, make sure to stop by the Late Breaking Results posters at 1 pm EST, where my amazing co-authors will be presenting our work!

🔗 Paper, videos and new Opus 5.5 results: agenticgentamp.github.io
agenticgentamp.github.io Coding Agents for Generalized Task and Motion Planning Problems Coding agents synthesize reusable robot policies for generalized task and motion planning. Explore results, videos, Python programs, and probing scripts.
Matteo Merler @merler_m Sep 25
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)
View quoted post on X
0 replies 2 reposts 14 likes 1.7K views View on X
Matteo Merler @merler_m
We’re releasing every program behind our results, with transcripts and git histories.
@WangRobin58334 evaluated almost 30,000 commits on the held-out instances, so this can be a resource to understand how agents develop strategies for physical problems.
🤗huggingface.co/datasets/prpl-…
huggingface.co prpl-group/agentic-gentamp-runs · Datasets at Hugging Face Dataset with 3.16M rows (Tabular, Text). Explore, preview and download it on Hugging Face.
Matteo Merler @merler_m Sep 25
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)
View quoted post on X
Matteo Merler @merler_m
Interesting ideas from Tom as usual, it's somewhat crazy to think of looking up papers the same as books and games, but if it works for media why not try something similar.
Having some real incentives for reviewers (regardless of system) is the most important part to figure out!
Tom Silver @tomssilver Nov 7
A Case for Paper Grading Boards (Or: A Plea for More Proposals to Radically Reform Peer Review)

tomsilver.github.io/blog/2026/reth…
View quoted post on X
Matteo Merler @merler_m
Super exciting work by Yichao and the team at Basis! Turns out coding agents are not just great at writing policies, but at understanding how the world works through code as well. Excited to see what else they can do!
Yichao Liang @yichao_liang Oct 2
Agentic real-to-sim is getting increasingly good at rebuilding scenes in a physics simulator. But looking at a scene does not reveal how glue cures, how water heats, or how a fan's wind pushes things. And what if the simulator cannot model these mechanisms at all?

We present EMPIRIC, an agent that learns a residual world model: a physics engine extended with code for the missing mechanisms.

It solves all 25 runs across five simulated domains, where coding-agent baselines solve 14 to 16, and on a real robot it learns wind and domino masses from two gusts.

basisresearch.github.io/empiric/

[1/9]
View quoted post on X