Shuran Song

This account is verified. Learn more

@SongShuran

Assistant Professor @Stanford University working on #Robotics #AI #ComputerVision

California shurans.github.io Joined July 2016

569 Following 17K Followers 460 Posts

Latest posts 5

Pinned post

Shuran Song @SongShuran Nov 4

The science of pretraining deserves more attention!

What makes a good pretrained model? How to evaluate the it? When should we stop training? And how do we optimize pretraining for downstream SFT or RL?

Some answers may be counterintuitive, e.g., the best-performing checkpoint may not be the best starting point for downstream learning.

If you're curious now, check out @s_zhanyi's work “Beyond Validation Loss” 👇

Zhanyi Sun @ CoRL Verified account @s_zhanyi Nov 3

Can more pretraining improve a robot policy’s success rate, yet makes it worse for porttraining? 🤔

In our new work, we find that it can. Continued supervised pretraining of generative robot policies can improve reliability at the expense of coverage, and this tradeoff affects downstream posttraining.

Read on to see why this happens and how we can preserve coverage. 1/n 🧵

View quoted post on X

Shuran Song @SongShuran Oct 13

Every robot policy should be a 🕒 Chronos Policy
— one that explicitly models physical time across the entire system stack, from observation to inference to action execution.
Because the physical world runs on real time, not sequence indices.

Changhao Wang Verified account @ChanghaoWang78 Oct 13

Most robot policies treat time implicitly, using sequence position as a proxy for physical time. But real robot systems rarely run on a fixed clock. Sensors run at different rates, observations arrive with delays, and actions take time to compute and execute. Successful control depends not only on what the robot observes but also on the timing of observations and actions.

We introduce 🕒 Chronos Policy, a diffusion policy that explicitly conditions on observation capture times and requested action execution times. We use time-aware positional encodings to make attention depend on elapsed physical time. Because timestamps let the policy interpret each sequence's timing, we can randomize sensing and action timing during training without making those variations ambiguous. This exposes the policy to a wider range of timing conditions, improving robustness to timing changes at deployment.

View quoted post on X

Shuran Song @SongShuran Oct 8

The best robot training data may be the data that was never explicitly demonstrated!

Check out how NEEDLEWORK 🪡 stitches together suboptimal and failure trajectories to improve both demonstration quality and diversity.

Juntao Ren @JuntaoRen Oct 7

How can we train better behavior from imperfect robot demos?

In NEEDLEWORK, we study how we can get stitching to work on robot data, and more importantly, why these connections help.

Website: needle-work.github.io
Paper: arxiv.org/abs/2610.02339

View quoted post on X

Shuran Song @SongShuran Oct 7

It’s really surprising to me that such a small change in the action exploration space can make such a big difference for dexterous manipulation -- across benchmarks, tasks, hands, and even different learning or optimization algorithms.

@harshgupta tested it on so many settings, and it seems to consistently help. Maybe you try it too!

Harsh Gupta Verified account @HarshGupta Oct 7

Per-joint Gaussian noise is a terrible way to explore a 20+ DoF hand

We introduce EigenDEXplore, which explores along coordinated patterns from human hand motion, while keeping full control of every joint

Better learning, from tool use 🔨 to reorientation 🔄 to bimanual 👐

View quoted post on X

Shuran Song @SongShuran Oct 7

Really nice blog on the design trade-offs of UMI!

Gino Kreis @gino_kreis Oct 6

I've been building my own UMI gripper for the past month. Papers usually skip that initial phase, so I put everything in a blog post. ginokreis.com/blog/dream-umi

Unsure whether to use ArUco tags or an encoder? Unsure how to calibrate the VR controller? Have a read!

View quoted post on X