Did a model secretly train on another model's outputs? Providers can catch this from server-side logs — Anthropic recently reported catching 29M such exchanges. But those methods need usage logs and infrastructure access; they don't work from the model itself. Our paper does: compare a model to an earlier checkpoint of itself, and ask which candidate teacher best explains the shift between them. The true teacher stands out. 🧵

If you want a robot to do something well, you need to know how to talk to it. If you don't, you can learn, with Semantic Action RL! In our paper, @JagdeepBhatia8, @ajwagenmaker, @verityw_ show how RL over VLA prompts enables new tasks and learns blazing fast in the real world!

How can generalist policies adapt to new challenges at deployment using skills they already have? We optimize VLA *prompt inputs* with reinforcement learning, enabling efficient real-robot adaptation on complex tasks where existing methods struggle. 🧵 semantic-action-rl.github.io

✈️I'll be at ACL in San Diego this week! I'll be at the NLP CSS workshop on Jul 3 (on the 12pm panel) and we have two main papers: - aclanthology.org/2026.acl-lo…, Jul 5, 2-3:30pm poster (I'll be there) - aclanthology.org/2026.acl-lo…, Jul 7, 9-10:30am poster (@stefkrsteski presenting)

How can we scale perception-based humanoid learning without collecting massive humanoid teleoperation data? 🚀 Excited to finally share VLK! What excites me most about VLK is that it reframes data collection as a data generation problem. Instead of relying on expensive humanoid teleoperation, we automatically generate synchronized vision, language, and whole-body kinematics from reconstructed real-world scenes. Making this vision a reality required bridging three fundamental challenges: 👀 Perception: Bridging the RGB sim → real gap through visual domain randomization and motion blur mitigation during both training and deployment. 🤖 Embodiment: Bridging the kinematics → dynamics gap with real-time VLA deployment, test-time RTC, and SceneBot, enabling seamless deployment on a real humanoid. 🌍 Environment: Bridging the real-world → synthetic gap to enable scalable Vision-Language-Kinematics data generation through scene reconstruction and interaction synthesis. It has been an amazing journey working with such an incredible team. For a complete walkthrough of the project, check out @jiaman01's thread below 👇 🌐 Project: vision-language-kinematics.g… 📄 Paper: arxiv.org/abs/2606.30645 🎦 Video: youtu.be/ZB6k_iMJP7M Huge thanks to my amazing collaborators @jiaman01 @eric_srchen @TakaraTruong @ Pei Xu, and to our advisors @pabbeel @rocky_duan @KoushilSreenath @akanazawa @carlo_sferrazza @GuanyaShi @ckarenliu.

We could not be prouder of this years BAIR PhD graduates! Follow their journey to see where they go next @baifeng_shi @sea_snell @databoydg @enfleisig @graceluo_ @zhuhl98 @HaozhiQ @jdzamfi @LianJiachen @joshminwookang @junhaobearxiong @kaylolittlejohn @KentKChang @kvablack @KunheYang @lisabdunlap @LongTonyLian @maulik__bhatt @michaelpsenka @nathanlichtle @neerjathakkar @nikita_mehandru @NiklasLauffer @qiyang_li @vbenara05 @vonekels @xiuyu_l @Yiheng_Li_Cal @ZheFu1 Sampada Deglurkar Wei-Jer Chang Yichen Xie Yigit Erginbas bair.berkeley.edu/blog/2026/…

🚨 Learning from Situated and Embodied Workshop @ COLM 2026 deadline is today! Submit your work on agents, embodied AI, human-AI interaction! Non-archival; preprints and under-review work welcome. learning-situated-interactio…

Introducing WARP-RM: A Warp-Augmented Relative Progress Reward Model for Data Curation 🧵 We gave our t-shirt folding robot more demonstrations and it got worse. Every extra demo ended in a successfully folded shirt. The data wasn't bad. It was noisy. The policy couldn't tell productive motion from dead time, and it imitated both equally. So which moments of a demo are actually worth copying? 🌐 Project Website: uynitsuj.github.io/warp-rm 📄 Paper: arxiv.org/abs/2606.28320 💻 Code: github.com/uynitsuj/WARP-RM 📨 XDOF blog post: xdof.ai/blog/warp-rm

1/N Neuroscience and social science research on humans has shown: – Similar brain activity predicts friendship and cooperation – Diverse minds drive innovation We wondered whether AI-AI interaction would show the same pattern. It does. LLMs with similar internal representations cooperate more, but produce less novel output. 🧵 (ICML 2026)

We can learn a model that provides shaped "process rewards" for robotic RL, that evolves automatically as the policy gets better. This improves performance on benchmarks, and works in the real world! Some fun new work with Raymond Tsao & @ajwagenmaker

BAIR (@berkeley_ai ) Faculty Michael I. Jordan gave the Commencement Speech at this year's @BerkeleyCDSS graduation where he spoke on the theme "Human Connection in the Era of Internet Technology". youtube.com/watch?v=omEtpo3o…

We are excited to announce what we have been working on for more than six months: The OpenThoughts-Agent dataset and OpenThinker-agent models. More than 100 ablations on data curation for RL environments for coding agents. Our data recipe is SOTA over all open-data agents in their class. We post-train a Qwen-3-32B to get 26% on Terminal Bench and open all our training sets, data pipelines, experiments and models. Some lessons we learned for training agents vs reasoning: 1. The Diversity of tasks matters more, compared to reasoning (OpenThoughts-Agent vs OpenThoughts). You could teach reasoning from math and it transfered widely but RL environments seem to teach more specific capabilities, so each domain must be covered. 2. Filtering high quality and hard questions remains very important. (Was also true for OpenThoughts reasoning). We discuss several ways of filtering. 3. Synthetic re-writing and task augmentation didn’t give significant benefits in our experiments. Sampling multiple teacher rollouts per task did work (was also true for reasoning). Even when keeping the dataset size fixed, multiple answers gave benefits. The Multiple answers mystery is still valid for agentic environments. 4. Stronger models are not necessarily better teachers (was also true for reasoning). The stronger teacher for Quen-3 was GLM-4.7-AWQ and the Terminus2 harness in Daytona. We are releasing 100k tasks and trajectories. 5.Benefits from GRPO remain limited and still on-going. I currently officially hate GRPO.

Why diffusion denoising-based generative methods do not suffer the curse of dimensionality even though the data may lie in extremely high-dim spaces? Our new work, accepted by the JMLR: arxiv.org/abs/2409.02426 reveals the not-so-surprising secret: as long as the intrinsic dimension of the distribution is very low, the generative process can be extremely efficient and effective! It seems that a mixture of low-rank Gaussians is a universal model for all informative real-world data. as we stipulated in a former textbook of mine: Generalized Principal Component Analysis: vision.jhu.edu/gpca/, published exactly ten years ago!

Great work using offline agentic exploration to develop robot skills!

Excited to share T-Rex: Tactile-Reactive Dexterous Manipulation 🦖🤖 Touch is fundamental to human dexterity, yet most Vision-Language-Action (VLA) models either ignore tactile feedback or lack the ability to react to high-frequency contact signals. In this work, we tackle both the data and architectural challenges of tactile-reactive dexterous manipulation. 🦖 A 100-hour tactile-synchronized dexterous manipulation dataset with 7,700 trajectories, 22 motor primitives, and 200 everyday objects. 🦖 A tactile-reactive MoT architecture with spatial-temporal tactile encoding and asynchronous high-frequency tactile refinement. 🦖 A scalable training recipe combining 22,889 hours of human egocentric pretraining with tactile-grounded robot mid-training. Across 12 real-world contact-rich manipulation tasks, T-Rex achieves over 30% higher average success rate than the strongest baseline. We are fully open-sourcing the dataset, models, teleoperation stack, training code, and inference pipeline. 🌐 Project: tactile-rex.github.io/ 📄 Paper: arxiv.org/abs/2606.17055 💻 Code: github.com/ZhuoyangLiu2005/T… 🤗 Dataset: huggingface.co/datasets/zeka… 🧵 Thread ↓

Check our new work: T-Rex: Tactile-Reactive Dexterous Manipulation 🦖🔥 Touch is fundamental, yet most VLAs are too slow for high-frequency contact. We built a unified foundation model to change that, pushing the limits of the @DexmateAI robot and @SharpaRobotics hands. What’s inside: 🧠 Architecture: tactile-reactive MoT for async high-frequency refinement. 📦 Data: 100-hr tactile dataset: 7.7k trajectories, 200 objects. 🏋️ Training: 22.8k hrs human pre-training --> tactile mid-training --> high-frequency control 📈 Result: 30% success across 12 real-world contact-rich tasks. 🔓 Open-sourcing models, data, teleop stack, and code. tactile-rex.github.io

Children learn from play. Can robots do the same? We propose 𝐏𝐥𝐚𝐲𝐟𝐮𝐥 𝐀𝐠𝐞𝐧𝐭𝐢𝐜 𝐑𝐨𝐛𝐨𝐭 𝐋𝐞𝐚𝐫𝐧𝐢𝐧𝐠, a paradigm that gives embodied coding agents a play stage before downstream tasks arrive, and instantiate it with 𝐑𝐀𝐓𝐬 (Robotics Agent Teams), where robots discover reusable skills through curious play. Co-led with @jiaxin_ge_

Remember the backlash against all AI voices being feminine—Siri, Alexa, etc.? Now, companies use a wider range of AI voices with ambiguous names like “alloy” and “shimmer.” Bias solved forever—ha, no. What happens when people hear these supposedly neutral voices?👇

In computer vision, the field advanced significantly due to openly available datasets and models; hope this will be true in robotics as well.

We can convert human videos to robot hand-object interaction trajectories in 4D. Enjoy! Paper: arxiv.org/abs/2606.19333 Website: do-as-i-do.com Code: github.com/malik-group/do-as… Authors:@bhawna_paliwal_,@HarithejaE,@willjhliang, @pabbeel , @notmahi , @JitendraMalikCV