Pedro Lopes

I get a bunch of emails/comments about my single-take videos we do to showcase our lab's work, so we gathered them in one page: lab.plopes.org/videos.html (including an annotated history of how we started making these, and how they transformed over the years, to become more improv!)

@avizurlo For human objects yes, for some other objects other animal hands might be better grippers!

The terminal state of UMI grippers are human hands

@jacyanthis @GoogleDeepMind Go Jacy (and see you at CHI!)

Thrilled to be joining @GoogleDeepMind as a student researcher in SF! We're building a multi-agent system to scale AI safety research and ensure pluralistic alignment. I think this is a crucial piece of safe AGI development for cooperation across many diverse human and AI agents!

@BramVDBorght @Melexis_Sensors @brubotics Love this! Great tutorial.

🤖 Bringing touch to robotics! Join our #ICRA2026 tutorial to build a tactile gripper with the @Melexis_Sensors Tactaxis® sensor. youtu.be/xYrV3yiFp_M. Limited spots! 🚀 mech.vub.ac.be/multibody/ICRA… @brubotics

@_stc Looks nice!

plexi mounting day #wip

@AndrewAsksHow I suspect a lot of CS labs would buy it. Consumers... Maybe unlikely?

Would you buy this for $4,700

@kanair I thought the question was perfectly posed (despite some weird answers you got). It's a really good one, simple and deep.

I asked this question because it's now common to talk with non-human agents (i.e. AI), and I thought maybe we can also talk with animals if there is a technological solution for a better inter-species interface.

@shahdhruv_ @corl_conf Amazing, good luck with CORL! Hoping to attend soon

I am excited to serve as a Sponsorship Chair for the upcoming @corl_conf in Austin TX: the largest and most talent dense gathering of robot learning researchers ever! 🤠 Please reach out to sponsorship@corl.org for details re. booths, demos, talent program, recruiting, ...

@KDharmarajan123 Fantastic! I wonder if we can use this for human execution too? I.e., via embodied-ai.tech for instance. Maybe we'll try!

Check out our latest Dream2Flow work, which uses 3D object flow from generated videos to perform manipulation!

@lauriewired Awesome explanations as always! Indeed snares and most drum sounds are terrible through this algorithm

Telephone Hold music sounds really bad. Mostly because it’s mapping complex instruments to a human throat. Many phone lines these days use CELP, Code-Excited Linear Prediction algorithms. Music breaks down in weird ways when you turn a piano into…speech.

@murloren @ylecun @AdrienBardes Very cool congrats!!

I am very happy to share the result of my internship at FAIR (Meta): V-JEPA 2.1: Unlocking Dense Features in Video Self-Supervised Learning with @ylecun @AdrienBardes Our approach learns dense, spatially coherent features from video while preserving strong global understanding

@gabriberton Indeed, they literally agreed to not use LLMs and then .. used it

ICML found a very smart way to detect highly negligent reviewers. Desk rejected their papers And yet I saw posts complaining about it??? These are reviewers who fed the paper to an LLM and copy pasted its output, without reading it, despite agreeing not to use LLMs! I wish ICML named these reviewers so I'd know who to never choose as co-author. Desk rejecting is too kind

New AI tool/system that leverages vision models and helps planning physical movements: not with a robot, but by means of stimulating the body with electrical muscle stimulation to perform challenging physical tasks! Read more below!

Next Thursday, March 26, we will host a Special Seminar with Dr. Angelos Barmpoutis from the University of Florida. Join us! “Building Real-World AI Infrastructure for Diagnosis in Parkinsonism: From Multisite Clinical Data Collection to Commercial and Regulatory Readiness”

@artjng Was a great talk!

Excited to share my final PhD paper, "HapticPipette," which I presented remotely at #AugmentedHumans2026 this week! While my previous research focused on haptic rendering and generation, this project focuses on their fundamental issue: the "measurement bottleneck." The success of large vision and audio models is built on collecting massive amounts of real-world data via ubiquitous cameras and microphones. In contrast, acquiring physical properties like compliance (the non-linear force-displacement relationship that strongly contributes to perceived softness) requires active contact and precise recording. Because measuring this macroscopic deformation typically relies on bulky instruments, we lack the mobile tools needed to scale real-world haptic data collection. HapticPipette introduces an "egocentric force-displacement measurement" approach that turns the everyday action of pressing an object into a sensing interaction. By combining a thin, flexible finger-worn force sensor with a head-worn RGB-D camera, even novice users can capture the complex non-linear compliance of everyday objects entirely in situ, as shown in our studies. Crucially, this setup is well compatible with other natural finger interactions, such as existing MR hand gestures. As AI advances in haptics (like my prior work, Scene2Hap), scaling real-world data collection is the next major bottleneck. Given its simplicity, our approach could eventually enable in-situ robotic data collection to truly scale this process, unlocking both realistic haptic generation and broader physical AI capabilities, such as training robotic hands to grasp non-rigid objects effectively. A huge shoutout to my co-authors, Jianhui Yan, Maja Fehlberg, Roland Bennewitz, and my advisor, Jürgen Steimle!

"What if AI could show you how to perform a physical task through your own body?" I am excited to share our latest #CHI2026 Best Paper where we demonstrate an Embodied-AI for physical tasks using muscle stimulation 🔗 embodied-ai.tech @yun_yh_ @plopesresearch

Text or video is not only form AI can take. In embodied-ai.tech (#CHI2026, Best Paper 🏆) we create an embodied AI that acts via muscle stimulation to perform physical tasks: e.g., place a bike on a bus rack and more! youtube.com/watch?v=pJM2Z8… by @yun_yh_ @romainnith

@OshoDayo Was a great talk and paper. Congratulations once more!

AHs2026にて，”Induced Human-Driven Actuation: Controlling Timing of Human Movement via Constraint-Release Mechanism”がBEST PAPER AWARDをいただきました．ありがとうございます！そして鈴木くん，おめでとう！ > augmented-humans.org #AHs2026

Thank you very much for your participation in #AugmentedHumans2026 / #AHs26 !! We added the next place, awards, and a glance at the website. Congrats everyone!! augmented-humans.org

The highlight of #AugmentedHumans2026 was definitely Masa's incredible performance while Djing and Vjing using his #BCI, eye tracking devices and robotic arms! Super energetic performance and great panel at the end of the performance. #AHs2026 #AHs26