Associate Professor at @Berkeley_EECS, @berkeley_ai. KAIR, @nerfstudioteam. Amazon Scholar @ FAR. Prev advised @WonderDynamics and @LumaLabsAI. she/her.
Berkeley, CA
- Tweets 863
- Following 627
- Followers 22,255
- Likes 4,555
Pinned Tweet
So excited for this to finally be out:
A fully open-source data & stack released so we can learn the ABCs of bimanual manipulation together!
This was a huge team effort, and many long days and nights of work. Congrats to the amazing team, I'm very proud of you!
Introducing ABC: open data, training, and infrastructure for robotics.
We release the largest teleop dataset to date, and extensively investigate design decisions, pretraining, and post-training techniques.
@arthurallshire @Cinnabar233 @adamrasb @redstone_hong @davidrmcall
4
21
182
30,819
Seeing is an act in itself. We look at what matters, and so should robots!
EyeRobot 2.0 does this with gaze, enabling precise manipulation without wrist cameras. Key for tool use, where the tool can block a wrist camera!
Co-led by @justkerrding who'll be on the job market! Hire him :)
Almost all bimanual robots include wrist cameras. They add bulk, cost, and complexity, but are critical for performance. Are they really needed?
Excited to finally share EyeRobot 2.0, where we use active gaze from an ego stereo camera instead 🧵 (with @kushtimusprime)
1
2
48
6,284
Here’s a new ✨off-policy✨ addition to our series on RL for flow policies.
Idea is simple: hill-climb on ∇Q, filter for stability, flow-match the replay buffer. Sample efficient and works on humanoids, VLAs, and images!
Led by amazing @ChungMinKim, who’s on the job market!
Excited to introduce QF3: Fast Flow RL with Filtered Q-Gradients!
Train humanoids from scratch, fine-tune VLAs, even fine-tune image models... all with one simple off-policy update ✨
2
5
69
6,746
@ChungMinKim knocked this project out of the park! She asks the most insightful questions, genuinely cares about building useful things and doing what’s right, and makes amazing visualizations to communicate complicated ideas. Any team would be lucky to have her!
3
493
Since Skill from Video (2018) and Videomimic (2025) the unspoken challenge has been finding the right videos to teach robots at scale.
Zihan's work changes that with v2v, turning a few good demos into many counterfactual videos.
Another step towards scalable real2sim2real!
Humans generalize from rich interaction experience. We bring this to humanoid loco-manipulation through scalable real-to-sim-to-real.
Introducing PRISM, we expand 4 real videos into 256 counterfactual variants, recover robot-object trajectories, and train one policy that generalizes across diverse objects, scales, and spatial layouts.
CoRL’26 🚀 Website for details 👉 prism-real2sim2real.github.i…
Work done at Amazon FAR w/ @zhenkirito123 @pabbeel @rocky_duan @JitendraMalikCV @carlo_sferrazza @ckarenliu @GuanyaShi @akanazawa
1
7
62
18,745
If you’re at IROS2026 @Cinnabar233 is giving a talk about ABC at the open robotics software tomorrow 10/1 at 9:35am, room 302!!
sites.google.com/rice.edu/ir…
ABC full release! Full stack of goodies in here.
Everything you need to get started.
Here's a challenge: Can you get the robot to fold a paper airplane? Excited to see what cool things you come up with!
3
24
4,514
ABC full release! Full stack of goodies in here.
Everything you need to get started.
Here's a challenge: Can you get the robot to fold a paper airplane? Excited to see what cool things you come up with!
ABC is fully out and headed to CoRL 2026!
ABC-130K is the largest open teleop dataset to date: 3,500 hours, 130K+ episodes, 195 tasks, collected on an $8K bimanual setup. The full stack is open: hardware, training code, sim, and eval, so academic labs can do this research on equal footing.
Beyond the data, we ran the baseline science for the community to stack on:
• Sim-to-real correlation (r = 0.91 on task progress)
• Which offline metrics matter?
• Scaling law in compute (training steps and batchsize)
• Conditioning the policy
Project page: abc.bot/
Github: github.com/amazon-far/abc
1
2
57
12,511
A few years ago, it was hard to imagine OpenAI putting serious resources into 3D/4D.
Now large-scale real2sim is within reach!
These systems build on the tools we’ve built and what we’ve learned. Progress!!
So much more to do, and a whole new set of problems to explore.
3
11
254
57,696
We got 99% on ARC-AGI-3 by instilling analysis-by-synthesis into AI models.
[Schema] is basically VIGA squared: do inverse graphics, then do it again one level up — inverse dynamics.
Super simple, surprisingly effective! Analysis by synthesis FTW!!!
Today, we’re introducing [schema]: a harness reaching 99% RHAE with Opus 4.8 + Fable 5 and 95.35% with GPT-5.6 Sol on ARC-AGI-3 Public set.
[schema] makes an LLM think like a physicist. 🧵
10
13
215
36,768
Babies learn by being naturally curious. How do we get autonomous agents to do the same?
We revisited curiosity in 3D exploration and found that memory is key. This project taught me a lot about what kind of functions an agent and a "world model" need to have for this direction
🚀 🚀 🚀 Excited to share our new paper:
Remember to be Curious: Episodic Context and Persistent Worlds for 3D Exploration
What does it take for an agent to stay curious in a 3D world?
The answer is memory.
🌐 Project: recuriosity.github.io/
📄 Paper: arxiv.org/abs/2605.22814
💻 Code: github.com/recuriosity/recur…
2
20
147
34,427
P.S. @lily_goli is an absolute star! She is amazing and on the job market now, reach out to her!!
1
5
1,978
Very excited to share this work @davidrmcall did with the fantastic NVIDIA Finland team last year. We have a surprisingly simple, but sample efficient way to post-train a flow model with RL.
We developed a simple, sample-efficient online RL technique for post-training image generation models. We see it as a possible steerable alternative to CFG, driven by any scalar reward, including human preference.
3
12
108
22,293
This natural interface + sample efficiency enables actual online, on-policy RL with human preferences. Not offline feedback like RLHF or DPO.
*You get to see the model change according to your feedback as you click!*
Here's the progression from David clicking for a few hours:
1
1
5
1,463
Code is already out! See blog + paper for more details:
Blog: mcallisterdavid.com/fdfo-blo…
Paper: arxiv.org/abs/2603.12893
Code: github.com/NVlabs/finite-dif…
1
4
747
Diffusion forcing is great for sequence modeling!
We've been working on social behavior prediction but nothing's worked this well.
Representation matters: We encode motion with discrete latents & model relative relationships in a neat way. It handles multiple tasks & people!
When people share a space, their movements become intertwined. Embodied agents need to understand these social dynamics to interact effectively.
Introducing MAGNet 🧲, a unified autoregressive diffusion forcing model for multi-agent motion generation that captures these interactions.
MAGNet is flexible: predict the future, fill in missing motion, or have people react to each other, all while naturally scaling to N>2 people and generating ultra-long motion sequences.
3
16
153
27,475
Exciting new project from FAR! A key factor in making this work was using a classic graphics algorithm called motion matching to generate synthetic training data 😉
The interactive demo is super insightful too, all humanoids papers should ship with one!!
Can humanoids perform agile, autonomous, long-horizon parkour—based on what they see in the world?
We present 𝗣𝗲𝗿𝗰𝗲𝗽𝘁𝗶𝘃𝗲 𝗛𝘂𝗺𝗮𝗻𝗼𝗶𝗱 𝗣𝗮𝗿𝗸𝗼𝘂𝗿 (𝗣𝗛𝗣): a framework that chains dynamic human skills using onboard depth perception for long-horizon traversal.
1/6
1
5
80
11,186