Videos are a promising data source for scaling manipulation; we just need to find the correct way to use them! Check out the new work on pretraining manipulation skills from hand-object trajectories parsed from online videos. The code is open-sourced and would be a great

Better. Faster. Stronger. Amazing work by Dingqi Daisy Zhang We design a model-informed domain randomization procedure and a BC-constrained RL approach that enables astonishing generalization. Works great up to 16X outside the training range.

Robot See, Robot Do allows you to teach a robot articulated manipulation with just your hands and a phone! RSRD imitates from 1) an object scan and 2) a human demonstration video, reconstructing 3D motion to plan a robot trajectory. robot-see-robot-do.github.io #CoRL2024 (Oral)

Welcome @Penn_ESE’s new Assistant Prof., Antonio Loquercio (@antonilog), who’s building robots that can sense & respond to human emotions. “For robots to be truly helpful, they must sense how a person is feeling and how they want—or don’t want—help.” bit.ly/48MNDiV

Happy to share our new work on Navigation World Models! 🔥🔥 Navigation is a fundamental skill of agents with visual-motor capabilities. We train a single World Model across multiple environments and diverse agent data. w/ Gaoyue Zhou, Danny Tran, trevordarrell and Yann LeCun.

We are making the Tesla Optimus walk increasingly more robust and ready for all terrains. Achieved by replacing a chunk of c++ code with nets!

I'm going to hire several PhD students this cycle. One on a specific topic (tinyurl.com/5yrpmeee), and others on blue-sky projects in robot learning. UPenn is a fantastic place to be (abundant robots and compute); come and look for yourself! Apply at tinyurl.com/3kxvbvvs

First class students (and teachers) :)

Thrilled to announce our #CVPR2025 workshop on Embodied Intelligence for Autonomous Systems on the Horizon featuring a crazy line-up of speakers and challenges. Mark it in your agendas and also in your registration #cvpr2025

Tired of the aggressive mediocrity of robust control (RC) and the unreliability of certainty equivalent (CE) control?! Then try domain randomization (DR)! We prove that DR-based control of an unknown linear system is nearly as efficient as CE control, and nearly as reliable as RC

Last day of classes for our course (antonilo.github.io/real_world_rob…) at Penn Electrical and Systems Engineering, co-taught with Dinesh Jayaraman. Congratulations to Zac Ravichandran and Ignacio Hounie for best project, and Lee Milburn for best paper presentation! Teaching this class has been a fantastic experience!

Our paper is accepted to T-RO IEEE Transactions on Robotics (T-RO)! Code for the learning algorithm is now open-source: t.ly/shgs9 Train your own extreme-adaptive policy today! 🚁🔥

We're continuing to push forward the idea of multi-sensory 3D simulators. I believe they will be very useful for AR/VR and (most importantly for me) robotics. If you want to know the key ideas behind it, come to our #CVPR2025 poster today (06/13) at 10:30 (#151).

I had a completely uneventful drive from an autonomous Wayve car in Nashville. It was amazing.

Thanks for coming to our Embodied Intelligence for Autonomous Systems workshop #CVPR2025 We had such a dense & fun program. Videos will be up soon. Kudos to the team: Chonghao Sima Ana-Maria M Christos Sakaridis Hongyang Li Jonah Philion Florent BARTOCCIONI Huijie Wang K. Chitta #cvpr2025

🤖 Does VLA models really listen to language instructions? Maybe not 👀 🚀 Introducing our RSS paper: CodeDiffuser -- using VLM-generated code to bridge the gap between **high-level language** and **low-level visuomotor policy** 🎮 Try the live demo: robopil.github.io/code-diffuser/ (1/9)