Jaden Clark

How can we leverage human video data to train generalist robot policies? 🤖 Enter RAD: Reasoning through Action-Free Data, a new way to train robot policies using both robot and human video data via action reasoning. rad-generalization.github.io

π, But Make It Fly ✈️ We fine-tuned π0, a VLA model pretrained entirely on manipulators, to fly a drone that picks up objects, navigates through gates, and composes both skills from language commands.

Can we learn whole-body mobile manipulation directly from human demonstrations? Introducing Whole-Body Mobile Manipulation Interface (HoMMI) Egocentric + UMI, 0 teleop -> bimanual & whole-body manipulation, long-horizon navigation, active perception hommi-robot.github.io

We find that RL post-training can substantially improve BC policies without teaching them anything fundamentally new. So what is RL doing? In DICE-RL, it contracts a broad behavior prior toward high-value modes. (1/n) zhanyisun.github.io/dice.rl.2026/

For video generation in robotic applications, looking pretty is usually not enough. Robot manipulation requires understanding how visual observations and 3D geometry evolve over time under agent actions, with temporal coherence and geometric consistency across camera views. We study this challenge in our work (recently accepted by @iclr_conf ), 4D Video Generation for Robot Manipulation, which enforces multi-view 3D consistency via geometric supervision to generate spatio-temporally aligned videos.

We release Cosmos Policy 💫: a state-of-the-art robot policy built on a video diffusion model backbone. - policy + world model + value function — in 1 model - no architectural changes to the base video model - SOTA in LIBERO (98.5%), RoboCasa (67.1%), & ALOHA tasks (93.6%) 🧵👇

A key aspect of scaling robot data collection is sensor reliability (hence why we haven't really seen tactile sensing at scale yet). UMI-FT addresses this by giving UMI robust/reliable fingertip level force-torque sensing, a major step toward scaling data for contact-rich tasks.

In Silicon Valley, “virtual cells” are suddenly everywhere. Meta and CZI recently went all in, signaling that this is no longer a fringe research direction. So what is virtual cell? Check out my new blog about virtual cell: yuncongtobymao.com/blog/virtual-c…

Simple heuristics can make a big difference when using foundation models for science. See our new paper on tracking animals with SAM 2, and check out some of the conservation work we're doing with it down in Costa Rica: woods.stanford.edu/news/pixels-pr…

Cool feature on some of the work we're doing out in Cambodia. Describing new species helps conserve key habitat in the region

Normally, changing robot policy behavior means changing its weights or relying on a goal-conditioned policy. What if there was another way? Check out DynaGuide, a novel policy steering approach that works on any pretrained diffusion policy. dynaguide.github.io 🧵

Thrilled to share that I'll be starting as an Assistant Professor at Georgia Tech (@ICatGT / @GTrobotics / @mlatgt) in Fall 2026. My lab will tackle problems in robot learning, multimodal ML, and interaction. I'm recruiting PhD students this next cycle – please apply/reach out!

Excited to join this cohort and continue my research in AI/robotics as a PhD student next fall!

Here is another uncut video of real-time interactions with @GoogleDeepMind 's Gemini Robotics!

(6/6) This was a super fun project with @JoeyHejna and @DorsaSadigh! Check out the full paper: lgpl-gaits.github.io

(5/6) We also conducted a user study where users specified their own arbitrary behaviors and ranked a set of trajectories. Users preferred lgpl 77% of the time!

What if I want my robot dog to act excited to see me? 🤖🐶 We introduce Language-Guided Preference Learning (LGPL), combining the efficiency of LLM parameterization with the precision of preference learning to generate 𝐞𝐱𝐩𝐫𝐞𝐬𝐬𝐢𝐯𝐞 behaviors. lgpl-gaits.github.io

Behavior cloning... clones behaviors, so naturally data quality directly affects performance. However, there aren't great ways of measuring how "good" or "bad" different demonstrations are. Our recent work seeks to address this problem using estimators of mutual information... 🧵