Vector Wang

Catching a flying ball is hard. What if with a flat plate? 🏓 Our work at RSS’26 shows it’s possible through Zero-Shot Sim2Real With Domain-Randomized Instance Set (DRIS), we catches different kinds of balls without any real-world fine-tuning 🔗 rice-robotpi-lab.github.io/DRIScatch/

@sentientcar this case it won’t make the full use of the soft finger?

The umi hand work continues!

@HelmickAlan Actually the first author of this paper has also participated in that work

@VectorWang2 I watched two Realarms juggling yesterday and I think that's even more difficult than catching a ball, catching and holding three in the air. Do juggling next. Excellent work Vector.

Catching a flying ball is hard. What if with a flat plate? 🏓 Our work at RSS’26 shows it’s possible through Zero-Shot Sim2Real With Domain-Randomized Instance Set (DRIS), we catches different kinds of balls without any real-world fine-tuning 🔗 rice-robotpi-lab.github.io/DRIScatch/

We put DRIS to the test with a highly demanding task: reactive catching using a completely flat plate on a Franka Robotics FR3 @FRANKAROBOTICS robot. Unlike robotic systems that rely on cups, nets, or enclosing surfaces for passive stabilization, our setup offers zero mechanical help, making it extremely sensitive to contact timing and noise.

Accepted at Robotics: Science and Systems (RSS Conference) 2026: "Zero-Shot Sim-to-Real Robot Learning: A Dexterous Manipulation Study on Reactive Catching." Authors:Kejia Ren¹, Gaotian Wang¹, Andrew S. Morgan², Kaiyu Hang¹ ¹Rice University ²Robotics and AI Institute

Same as ManiDreams, this task is also fully trained within Maniskills @Stone_Tao 😂😂

Here is how it works:  Instead of training on one randomized instance at a time, DRIS propagates multiple randomized instances simultaneously under a shared action.  By allowing the policy to learn from a set of possible dynamics outcomes at once, it becomes remarkably robust to variations in physical properties and sensing errors

We're building the UI @LeRobotHF should have shipped with Plug in your SO101 or XLeRobot, scan your ports and cameras, auto-calibrate, and you're teleoperated in under 5 minutes. No terminal.

@pkuhar In US the official price for the large arm is ~2k$, ~1.4k$ in China. I am building a smaller version myself using their cheaper motors.

@VectorWang2 Interesting, do you have a link? When I looked it up it was 2.5k

Building up a mini Tidybot for <$2k in total. More dynamic (realtime no speed up) and more robust Plugged into upcoming X-bot universe with OpenClaw agentic applications. github.com/TidyBot-Servic…

@pkuhar The arm is ～1k$ if assemble it yourself, all parts are open sourced

@VectorWang2 Makes sense. Currently the BOM looks like way over the $2k, but I could easily see this being sub $1k. That arm alone is $2k+

@kamathsblog Here's the step file of the wheel base github.com/Vector-Wangel/…

@VectorWang2 Are the 3D files for the hoverboard motors available? I ordered one of these trolleys to put on a mecanum wheeled robot, but I've also got spare hub motors

@pkuhar no, currently using an odrive 3.6 board

@VectorWang2 esc included?

@R3PL1C8R The chute can be just 3D printed, already have a version for both tables and floors.

@VectorWang2 Needs a round chute for the cans and water bottles to go into. Also needs to reach the floor in homes. Pick stuff up and drop it into its bin.

@pkuhar The cheapest Direct Drive scooter wheel. ~20$ cost, can be as fast as 20 mph

@VectorWang2 This is next level. What did you use for wheels?

@SGhashghaei in a more reliable way

@VectorWang2 What happens to all the humanoid robotics companies raising hundreds of millions when these robots are 10% of the price and do roughly the same thing???

@VMises76153 Stringman is an amazing design😄. And it could be useful to plug into Xbot universe. You can checkout the general framework first

@VectorWang2 I'd like to try to tie in your framework with Stringman at some point but it's a lot to think about.

Watch AthenaZero juggle barehanded using on-board sensory feedback only. No motion capture. No funnels. No help adding the third ball. The robot learns to adapt to the uncertainties from contact and the appropriate hand-eye coordination. Learn more: rai-inst.com/resources/blog…

@chris_j_paxton Why is this background similar to LingBot?

Sudo is training robots entirely in simulation and has shown their robot running for an uninterrupted 60 minutes

ManiDreams is not a standalone method, but: An Open-Source Library for Robust Object Manipulation via Uncertainty-aware Task-specific Intuitive Physics Authors: Gaotian Wang¹, Kejia Ren¹, Andrew S. Morgan², Kaiyu Hang¹ ¹Rice University ²Robotics and AI Institute (Video: Fast Foundation Stereo for perception + Newton for physics backend, real-time 3D DRIS physics prediction)

No world model is accurate, especially the intuitive one in your head Most video models collapse the future into one deterministic rollout, slowly ManiDreams keeps the uncertainty, and plans over it in a modular framework 😴Dream, 🤔Predict, 📦Constrain rice-robotpi-lab.github.io/ManiDreams/