


humans, robots, and a secret third thing.. @waldenrobotics


I had a great time seeing everyone at CoRL this week. My talk was called "Learning in the factories", and I talked about: - Walden's longest-running deployment to date, - the value of coupling deployment data with large-scale pretraining, and - a few things I've learned as I've broadened my focus from AI/robotics to include large-scale manufacturing. I'm excited to keep sharing more of Walden's story as we go. Things are moving very fast! - youtube.com/live/uPXoJm-L3YY…

I joined @WaldenRobotics just a couple weeks ago and the progress since then has been amazing. Feeling incredibly lucky to work with this team and excited for what’s to come!
At CoRL, @RussTedrake shared the work behind LBM 2.0: Sage and what we’re learning from deploying robots in factories. Watch the keynote: youtube.com/live/uPXoJm-L3YY… More technical detail in the blog: waldenrobotics.com/news/lbm-…

Robots in the wild should translate into valuable & diverse interaction data. But how can we surface these interactions, and learn from them? FIDO powers its data flywheel with play! Our intuitive & friendly robot dog learns a dynamics model 💭 while playing fetch with you 🎾🐕
What can a robot learn from playing fetch? 🐕🎾 Meet FIDO: a robot that turns casual play into real-world dynamics data. We collect data to fine-tune a VLM dynamics model that transfers to sports prediction and balloon catching tasks. 🔗 fido-robot.github.io/site 🧵1/5

I think the coolest part of FIDO (apart from the fact it has a massive magnetic tongue) is that the data is a side product of a normal interaction, and yet it’s so useful! As always, a pleasure to work with Michelle, Jenn, and everyone else!
What can a robot learn from playing fetch? 🐕🎾 Meet FIDO: a robot that turns casual play into real-world dynamics data. We collect data to fine-tune a VLM dynamics model that transfers to sports prediction and balloon catching tasks. 🔗 fido-robot.github.io/site 🧵1/5

Excited to share this step toward more intuitive interfaces for robot interaction! Had a lot of fun working on this with @du_maximilian, @yvsriram, @ajaysridhar0, @mazrk7, @DorsaSadigh , and @jenngrannen :) Paper & videos: fido-robot.github.io/site 🧵5/5

We also consider the task of robot balloon catching 🎈 Using a dynamics model fine-tuned with FIDO data yields a 20.0% improvement in balloon catching over baselines trained without FIDO data. 🧵4/5

With data across 18 users and 3 environments, we fine-tune a VLM to predict projectile object motion. The addition of FIDO data reduces prediction error on two unseen domain-adjacent sports tasks 🏸 ⚾ , and performance improves monotonically with diversity of data! 🧵3/5

Crowdsourcing diverse data often requires significant infrastructure or operator expertise. What if it were as easy as playing fetch? Fetch is easy to understand, engaging, and naturally repetitive, lending itself to in-the-wild collection in a variety of environments. 🧵2/5

What can a robot learn from playing fetch? 🐕🎾 Meet FIDO: a robot that turns casual play into real-world dynamics data. We collect data to fine-tune a VLM dynamics model that transfers to sports prediction and balloon catching tasks. 🔗 fido-robot.github.io/site 🧵1/5

it was so fun working on this last summer and i can't wait for others to see it!
Dear Friends, Years ago, standing deep in the Amazon rainforest with Efsun, I realized I was not observing nature — I was standing inside a living intelligence. The forest was computing its own environment, translating invisible forces into physical form. That experience changed everything. 🌿 For ten years, we painted with data. Light was our brush. Buildings were our canvas. The pigments never dried because they were never physical. But we never stopped thinking about the canvas — the real one. The one you can touch. For the past year, our studio has been quietly building a bridge. We connected our Large Nature Model to advanced robotics, and for the first time, we are translating the memory of the Earth and the emotions of our collectors into real, physical brushstrokes. Real canvas. Real pigment. Human + Machine collaboration. 🤖🎨 Today, I am proud to share QUALIA — the first physical data painting series by our studio. 365 unique editions. Private sale. See you at DATALAND. ✨

machinelov.ing might be the best domain ever ??

the six love languages are words of affirmation, quality time, gift giving, acts of service, physical touch, and.. ..machine learning. this year ML valentines have a new home on the internet <3

Erdem is an amazing mentor, consider applying!
Looking for multiple undergraduate interns to work on robotics research projects at USC this summer (@CSatUSC @USCViterbi). It is an NSF REU. US citizens & permanent residents are eligible. Info: cs.usc.edu/reu/ P.S.: We will extend the deadline by at least a week :)

stumbled upon A Journey to the Year 2025, a short story from 1922 that imagines a future of individual human flight, weather machines, and homes commanded by voice. it's always interesting to read what people of the past dreamed our world today would be.. flyingcarsandfoodpills.com/a…

Data collection remains a bottleneck in imitation learning for robotics: it’s tedious & often needs access to a robot. Can we make the data collection process more accessible and engaging? We introduce RoboCade, a platform for gamifying remote robot data collection 🎮🤖 (1/6)