Ivrolan

- Jean Dausset, Ciudad de las artes y las ciencias

@IntellRobotLabs buen título ! 🥳

Buenas semana de resultados de investigación!! Acaban de comunicarnos la aceptación de paper "I Move Therefore I Learn: Experience-Based Traversability in Outdoor Robotics" en IEEE Access (JCR Q2) tras muchos meses de revisiones. Gracias a Miguel por su perseverancia y trabajo 🤓

Tricks that work in the lab don't work at the World Cup! 🧵(1/5)

Introducing NEO’s 25 Degrees of Freedom, tendon-driven hands — nearing or surpassing human-level dexterity, strength, speed, and reliability. For seventy years, robotics worked around the hand problem. The humanoid bet is the reverse: it lives or dies at the fingertips.

long live robots with cute heads the eyelids to protect cameras + privacy + "sleeping" mode that people easily understand is top choice

and I was wondering why I was running out of space... there was a lot of stuff under the rug haha

It's amazing what you can do without machine learning. I've spent hours watching my simple implementation of the Dynamic Window Algorithm, basically a model-predictive control method. Runs live in the browser at this link: doc.ic.ac.uk/~ajd/Planning/

have to respect researchers with limited resources 🫡

𝗥𝗼𝗯𝗼𝘁𝘀 𝗱𝗼𝗻’𝘁 𝗻𝗲𝗲𝗱 𝗺𝗼𝗿𝗲 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻𝘀. 𝗧𝗵𝗲𝘆 𝗻𝗲𝗲𝗱 𝘁𝗼 𝗹𝗲𝗮𝗿𝗻 𝗳𝗿𝗼𝗺 𝗳𝗮𝗶𝗹𝘂𝗿𝗲 — 𝗮𝗳𝘁𝗲𝗿 𝘄𝗮𝘁𝗰𝗵𝗶𝗻𝗴 𝗵𝘂𝗺𝗮𝗻𝘀. Most robot learning systems assume failure is the end of learning. In our new work, we study whether robots can improve after deployment by learning from their own failures, without any human intervention, teleoperation, or corrective labels. The key idea is simple: human videos contain structure about how the world works. We use them to learn cross-embodiment representations of action, dynamics, and value, enabling a shared predictive space between human behavior and robot experience. This allows a new learning loop: 👉 pretrain on human videos 👉 deploy robot policy 👉 observe failures 👉 reinterpret failures using human priors 👉 improve autonomously We evaluate this across 7 real-world manipulation tasks, showing: 📈 40% → 81% success rate 🏆 Strong improvements over π0.6 RECAP and RISE ✔️ Zero human intervention during post-deployment improvement 🧬 Generalizes across robot embodiments and policy backbones A key finding is that explicit failure repair significantly outperforms failure reweighting, yielding substantially larger gains under identical data conditions (+25 pts vs +5 pts on the same π0.5 base policy). Overall, the results suggest a shift in how we think about robot learning: Human videos are not only for pretraining policies. They can provide the structure needed for continual self-improvement after deployment. 📄 Paper: arxiv.org/pdf/2606.21406 🌐 Project: ethz-mrl.github.io/robot-self-imp… I am grateful for working with the fantastic leads @hanzhic678 and @Anran_zh, and our collaborators Simon Schaefer, Kejia Chen, Shi Chen, Daniel Cremers. Special thanks to @StefanLeuteneg1 for co-advising this project with me. @ETH @TU_Muenchen @Microsoft Check out Hanzhi's 🧵 for more details x.com/hanzhic678/sta…

@Laz4rz @ETH_en @EPFL_en @cscsch this is great news! do you know any project that uses apertus in robotics?

Apertus is now also present on X! The @ETH_en, @EPFL_en and @cscsch co-initiative is one of the only ones in Europe that seem to have a real shot at building a lab culture similar to one from a proper frontier lab. There isn’t too much pressure on project-based results, instead there is a vision on compounding the learnings inside the team, iteratively producing better models and also sharing with the academic community

Rodrigo Pérez-Rodríguez y yo hemos creado una web para hacer accesible nuestro libro de Arquitecturas Software para Robots que hemos creado en el @IntellRobotLabs de la @urjc Incluye el libro y las traspas que hemos usado durante el curso. intelligentroboticslabs.github.io/web_libro_arqu…

IK alone can take you pretty far. I put together a really cute demo to demonstrate this. Watch this Panda robot CNC engrave the MuJoCo logo into a curved dome using mink.

Now that #IROS26 results are out, I would like to share a few thoughts of the peer-review process as AE. 1. If you submit, you must review at least 2 papers! Even if not entirely on your topic, you can still as a researcher look for sources and learn whether the work is novel...

@LouisLeLay4 the saddest part is no head nor eyes :( we have to stop minimalism here

calling it a humanoid with no legs, my heart is closed… really nice design though!

Awesome-UMI を公開しました！ 特にヒューマノイドロボットの領域でボトルネックと言われている人間のデモンストレーションデータを収集する装置、「UMI（Universal Manipulation Interface）」系の装置をまとめたウェブサイトです。 まだ完璧ではないベータ版ですが、継続的に追加できればと思います

Today, we bring SAC to RSL-RL, one of the most widely used RL frameworks in massively parallel robot learning, developed at RSL @leggedrobotics. We try to understand the long-standing performance gap between SAC and PPO, and crystallize important factors sabagian.github.io/sac_release_pr…

@WenCabrel_ @pesarlin there are ways to anonymize code repositories before publicly release it after acceptance, but that's not the issue here. The paper is now accepted with a project page and blog post, but still no code.

If I remember correctly, it was a double-blind review process, so authors should not have added their repo or anything else that would allow reviewers to identify the paper as theirs. I am sure this was done to eliminate favoritism. Once your paper is accepted, you can disclose anything—the code and so on.

Where is the code? No code, no public API, private datasets, too little details to reproduce = low impact. What does the award committee value?

#ICRA2026 parade of robots 🤖

@facontidavide WOW Plotjuggler will become the Everything App !

I could not resist. #Plotjugger secret cheat code!