Keivalya Pandya (in Bay Area)

founding mini-VLA | Robotics @GM |prev. @FamiliarMxM, @isro | 8x Hackathon 🏆

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Turns out, the hardest part of building a robotics platform wasn't writing the code.

It was learning what users actually needed.

Building future of robot simulation @ https://t.co/HOgks9lIRH

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Can someone please endorse me for cs[dot]RO on ArXiV?

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Now that is something 🤯😭

[The Humanoid Hub @TheHumanoidHub]

A stunning piece of engineering!!

1X unveils the new humanoid hand for NEO

- 25 degrees of freedom: 22 fully actuated in the fingers and palm, plus 3 at the wrist.
- The DoF are distributed anatomically rather than evenly, deliberately biased toward a thumb that genuinely opposes the fingers.
- In-house tendon-driven, quasi-direct-drive running low gear ratios of ~5:1 to 15:1 vs the typical 100:1–200:1.
- Motors live in the forearm and pull tendons through the wrist. This keeps the hand light and its inertia low while producing high forces.

Sensing

- All 25 DoF are natively force-controlled and fully backdrivable. Every joint doubles as a force sensor.
- Very important, closed-loop proprioception: it always knows its own pose and effort without looking.
- Tactile skin across the fingertips and surfaces measuring contact and shear. This helps with adaptive gripping in real time.

Safety and durability

- IP68 waterproof and food-safe, so it can wash its own hands.
- Compliant by construction: the low gear ratios, tendon drive, and low distal inertia let external impacts safely backdrive the fingers. It yields when hit by a hammer or caught in a drawer.
- Full finger assemblies validated to millions of cycles.

Manufacturing

- Deep vertical integration: in-house motors, custom electronics, and tendon systems.
- Hundreds built already, with capacity to produce 10,000 hands this year.

"An API to the Physical World"

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@lucyjcai Isn't that too expensive? It's definitely fire tho

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@RemiCadene @UMA_Robots Interesting. Trained on how many demonstrations?

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Data, not compute or models, is the #1 bottleneck for embodied AI -- physical trajectories don't scrape like internet text.

(reading) https://t.co/MVI6bpuXhh

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Today robotics gold rush isn’t in building another robot.
It’s in owning the data pipeline that teaches robots how the physical world actually works.

Hardware commoditizes. High-signal physical data does not.

[Keivalya Pandya (in Bay Area) @KeivalyaP]

ASPIRE (@nvidia) is a robot agent that learns how to debug & remember. interesting to see code-as-policy where people simply use claude-code/codex to control the robots!

(i am particularly impressed by the performance)

https://t.co/6DJ32yJI9n

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@xulux_ai @assistantui Pretty cool. I'd like have one for https://t.co/8LeUY3cPfi someday!

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@lucyjcai In my undergrad I used to prosthetic limbs (patented), some FIRST Robotics style robots, self-balancing, line followers, and more lol

Now, it's more into companionship and industrial robots, depending on which team I'm working with

Now I'm into WAMs @ GM.

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@lucyjcai Yesss! We do. I'll announce next meetup+paper soon. Stay tuned!

Btw, we cover VLAs and WAMs too!

[Keivalya Pandya (in Bay Area) @KeivalyaP]

starting a Robotics Research Reading Club -- will cover sota papers in robot learning
https://t.co/ysPc8iw650

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@buildanythingso I'm down!

[Keivalya Pandya (in Bay Area) @KeivalyaP]

Happy 250th, America 🇺🇸
Grateful for everything this country's given me so far. Just getting started.

[Keivalya Pandya (in Bay Area) @KeivalyaP]

If you could read just one paper from each area of Robotics, make it these:

• Motion Planning → RRT (Rapidly-exploring Random Trees)
• SLAM / Navigation → FastSLAM
• Imitation Learning → DAgger
• Sim-to-Real Transfer → Domain Randomization
• Robotic Grasping → Dex-Net
• Continuous Control → Soft Actor-Critic (SAC)
• Legged Locomotion → Learning to Walk via Deep RL (ANYmal)
• Vision-Language-Action → RT-1 (Robotics Transformer)

[Keivalya Pandya (in Bay Area) @KeivalyaP]

This Transformers refresher is fantastic! I highly recommend it to beginners as well.

https://t.co/gby8c0K6MD

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@dylanmatt I wouldn't get it, just yet

[Ronit Pereira @CAronitpereira]

“It's very hard to fail completely, if you aim high enough.”

- Larry Page

[Keivalya Pandya (in Bay Area) @KeivalyaP]

If you want to actually understand VLAs (not just run someone's checkpoint), read these in order:

1. RT-1 (transformers can drive a robot)
2. RT-2 (the paper that coined "VLA" — web knowledge → actions as tokens)
3. Open X-Embodiment (data diversity beats one perfect robot)
4. OpenVLA (a 7B VLA you can actually fine-tune)
5. π0 (flow-matching action expert, the current frontier)

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@_krishnasuresh This is so cool 🤯

[Keivalya Pandya (in Bay Area) @KeivalyaP]

@SrushtiMad @ycombinator @hud_evals thankyou! and i gotta prove that 😭

[Keivalya Pandya (in Bay Area) @KeivalyaP]

thank you @ycombinator and @hud_evals for beautiful memories. i'll be back -- i'm insane
~ team Atomz