David Fang

Building | Prev: @DynaRobotics, @Cruise, @UMich @sjtu1896

[David Fang @dlfangda1]

Would be really exciting to mid-train a robotics model from Astra. The strongest foundation models are starting to look increasingly robotics-native out of the box.

[Jay Chooi @chooi_jeq]

GPT-6 Astra scored 95% on a robot control task, up from Fable 5.1's 40%, with 6.2x fewer output tokens at 2.3x lower cost. 🧵

[David Fang @dlfangda1]

Traditional robotics is good at executing predefined tasks.

Physical AI today is getting good at executing structured workflows with flexibility and variation within the steps.

But a lot of the most valuable physical work looks different: new situations, new workflows, new structures, with little or no prior demonstration.

We’re still pretty far from that.

[David Fang @dlfangda1]

Love to see the team keep pushing the frontier. Learned something new about unblocking scaling laws with human data from the deep analysis here. Thanks for sharing so openly!

[Dyna Robotics @DynaRobotics]

Today we are introducing Dyna-2, a world-action model pre-trained on one million hours of human video. At this scale, for the first time, we discovered several new scaling laws:

• world-action models exhibit scaling law on human data across four orders of magnitude, from 1000 to 1,000,000 hours,
• this human data scaling law implied a scaling law on never seen robot data,
• both data and objective matter; world modeling and scaling on video data are essential for cross-embodiment scaling transfer to emerge

🧵

[David Fang @dlfangda1]

One thing I’m learning while exploring robotics:

There are endless problems robots could solve.

The hard part is finding the ones painful enough, repetitive enough, and valuable enough that someone actually wants the robot today.

[David Fang @dlfangda1]

@kun_h____ MVP

[David Fang @dlfangda1]

@JasonMa2020 @YorkYang5050 1, 2, 3 Dyna-2!

[David Fang @dlfangda1]

The World Cup 2026 schedule is honestly confusing when it comes to figuring out which teams would meet each other in the knockout stage, and I ended up buying the wrong ticket by mistake lol

So we build a prediction site with agent:
pick group results, simulate the knockout bracket, see which teams face each other along the way, crown a champion, and share your prediction.

Click “info” on any simulated matchup to see mainstream ticket links for that game.
Check it out here: https://t.co/qPGl2XBx5K

#WorldCup2026

[David Fang @dlfangda1]

It’s wild how fast robotics software development is becoming with coding agents.

Started a hobby robot project from scratch, and in one night I had the data collection / training / inference / visualization / runtime stack running and got the policy working.

But robotics is still hard for the same reason it has always been hard: the physical world.

Most of the time went into assembling the SO-ARM101 and other components, waiting for 3D-printed parts, collecting real robot data, and dealing with hardware details.

[Chris Paxton @chris_j_paxton]

Again robotics software sucks. Its so bad. Part of what makes it so bad -- and what makes stuft like ROS useful -- is the amount of boring middleware code you need to write between sensors, processes, etc. Coding models make it so much easier to work with, its wonderful.

[David Fang @dlfangda1]

@gs_ai_ Impressive dexterity. I’d love to see this extend beyond staged tabletop long-horizon tasks into real human workflows, where robots move around, adapt, and solve problems end-to-end.

[Stephen James @stepjamUK]

𝗦𝗵𝗶𝗿𝘁 𝗳𝗼𝗹𝗱𝗶𝗻𝗴: 𝟳𝟱.𝟯 𝘀𝗲𝗰𝗼𝗻𝗱𝘀 𝗶𝗻 𝗱𝗲𝗺𝗼𝗻𝘀𝘁𝗿𝗮𝘁𝗶𝗼𝗻, 𝟭𝟴.𝟵 𝗶𝗻 𝗲𝘅𝗲𝗰𝘂𝘁𝗶𝗼𝗻. 𝗦𝗮𝗺𝗲 𝗽𝗼𝗹𝗶𝗰𝘆. 𝗡𝗼 𝗿𝗲𝘁𝗿𝗮𝗶𝗻𝗶𝗻𝗴.

The instinct, when a robot runs slowly, is to reach for a better model. Usually that isn't where the problem is.

Assuming you already have a trained policy (which is its own unsolved problem, and not one we should pretend is behind us), teleoperation data is captured at human speed because hardware limits and delayed control feedback force it to be. The policy learns the behaviour. It also learns the tempo. Naive acceleration breaks the system long before it breaks the model.

The real bottleneck sits in the latency stack. Cameras capture frames 55ms before the timestamp. Joint position readings arrive 50ms late. Tracking controllers lag by 150ms. These delays compound during execution, producing a gap between what the model expects and what the hardware actually does. Push the speed without compensating, and the trajectory goes jerky, the arm oscillates, and success rates collapse.

@Dexmal_AI's Realtime-VLA V2 paper treats this as a system problem rather than a model problem. Three layers sit around the unchanged policy: a calibration pass that measures and aligns the real delays against training data, trajectory post-processing that uses quadratic programming to smooth action chunks and pre-amplify commands against control loop lag, and a human-in-the-loop speed adaptation head that learns when to accelerate on low-risk segments and slow down for precision-critical ones.

The paper reports execution close to human tempo across physical tasks. PCB placement: 89.5s to 37.8s. Pick-and-latch: 98.6s to 42.6s. Shirt folding within a fraction of a second of the human operator. Real hardware, not simulation.

What I find useful about this isn't the raw numbers. It's where the optimisation happens. The policy is untouched. The win comes from infrastructure most teams treat as an afterthought: time calibration, trajectory smoothing, adaptive execution.

There's a broader point worth drawing out. The field spends a lot of attention on model architecture and much less on the synchronisation, calibration, and execution layers underneath. That imbalance is how teams end up debugging six months of system instability that has nothing to do with the model they trained. Data is still the harder problem, and I don't want to pretend otherwise. But once you have a policy worth running, whether or not it runs well is a systems question.

𝗠𝗼𝘀𝘁 𝗼𝗳 𝘄𝗵𝗮𝘁 𝗹𝗶𝗺𝗶𝘁𝘀 𝗿𝗼𝗯𝗼𝘁 𝗱𝗲𝗽𝗹𝗼𝘆𝗺𝗲𝗻𝘁 𝘁𝗼𝗱𝗮𝘆 𝗶𝘀𝗻'𝘁 𝘁𝗵𝗲 𝗺𝗼𝗱𝗲𝗹. 𝗜𝘁'𝘀 𝗲𝘃𝗲𝗿𝘆𝘁𝗵𝗶𝗻𝗴 𝗮𝗿𝗼𝘂𝗻𝗱 𝗶𝘁.

Paper and project page in comments ⬇️

#RobotLearning

[David Fang @dlfangda1]

@aaronistan Very cool!!

[David Fang @dlfangda1]

@jsuarez I don’t think the core issue is science vs engineering, it’s goal setting.
If the goal is a paper/demo, you bias toward novelty.
If it’s a product, you bias toward robustness, iteration, and value.
Science and engineering are just tools. Strong teams get the goal right first.

[David Fang @dlfangda1]

@jin_linyi Congrats!

[David Fang @dlfangda1]

@chris_j_paxton We’ve actually been doing this all along

[David Fang @dlfangda1]

We’re building robots that can think, plan, and act — developing general agentic reasoning for embodied AI. Our robots break down complex, real-world tasks into smaller subgoals, reason about the next best action, and execute with efficiency and dexterity. Join us!

[Dyna Robotics @DynaRobotics]

🎃 Halloween is coming. Our hardworking team is lining up for sweet treats, of course, served by Dynasaur!

DYNA VLA model now has robust agentic reasoning capability, allowing it to serve arbitrary combinations and counts of candies! Pure imitation learning can’t work given the combinatorially many possibilities.

No video edits. Uninterrupted, real-life, as always 🤖

Happy Halloween from DYNA!🍬

[David Fang @dlfangda1]

@eddybuild Very cool, how are you dealing with occluded hands?

[David Fang @dlfangda1]

I am thrilled to be part of the DYNA-1 effort, we are on the road to build a general approach to make robots master any human skill and actually deploy and commercialize it. Please reach out if you are passionate to deliver embodied AI to empower everyone, we’re hiring!

[Jason Ma @JasonMa2020]

Introducing Dynamism v1 (DYNA-1) by @DynaRobotics – the first robot foundation model built for round-the-clock, high-throughput dexterous autonomy.

Here is a time-lapse video of our model autonomously folding 850+ napkins in a span of 24 hours with
• 99.4% success rate — zero human intervention
• 60% human throughput speed
• 4.3/5 quality ratings (set by the client)

A thread on our motivation, insights and results: