Sankaet

@sourceryy @a16z @aleximm lidar DOES NOT have better performance in fog, snow, rain. It’s quite the opposite. It fires laser pulses to create precise 3D point clouds. So raindrops, snow, fog scatter those beams and create noise. Radar is helpful in those situations

.@a16z's @aleximm says Waymo is "significantly ahead" compared to Tesla FSD when it comes to safety: "We love Elon." "Critics would say that LiDAR provides more data and has better performance in adverse weather like fog, snow, rain." "We believe the Waymo approach today is clearly safer." " That said—if you look at FSD, and the publicly available data, they have shown significant improvement. The slope is promising."

our convo with @sankaet building humanoid robots to one day build a city on mars is live now. - why they focus on heavy industrial tasks (and how to build actuators) - their 2030 master plan - what the next gen of founders should work on today @foundation_robo

@louismorgner Thanks for coming by

great time shooting with @sankaet today. youtube video dropping soon - great lessons for young founders.

"If we actually want to increase our overall manufacturing output, and in case of the US, reindustrialize America, we need automation." @sankaet of CEO at Foundation Future Industries at Reind 2.0

@JacobZietek correct

*per captia a16z.news/p/robotics-nee…

Robotics has spent decades optimizing for research. Deployment requires a completely different kind of person: operators, industrialists, and outsiders the field typically ignores. There's a wave of people who want to build in robotics. The field doesn't know what to do with them. New essay, Robotics Needs Fewer Roboticists* below 👇

You mean VLAs. VLMs are like your pre-frontal cortex. They will be responsible for reasoning and memory at a high level. Combining them in a world model handicaps you on the data that can be used to train this. You want to be able to use text to also learn higher-order reasoning. World models are like your pre-motor cortex; they build an intuition around conditioned and unconditioned sequences to then plan actions and how to do them. You can learn unconditioned sequences by watching videos on the internet and conditioned ones by robot “free play”. If the cost of data was equal across text, images, video, and teleop, then all of this would be in an end-to-end model. But it’s not, so hierarchical makes the most sense—just like the human brain.

just how teleop is pretty much becoming outdated this year, VLMs as backbone will be completely outdated in 2027

@philliplord1 @foundation_robo Thanks for coming by 🙂

Just visited @foundation_robo’s office in SFO—mind blown! Met founder @sankaet (Sankaet Pathak), the visionary behind autonomous humanoid robots like Phantom that can handle ANY labor humanity needs—from glass manufacturing to future defense. What sets them apart? Blazing deployment speed (40+ units already, scaling to thousands fast), true AI-native Physics Action Model for real autonomy, and aggressive real-world pilots in auto & consumer goods. No other robotics play is executing this hard. Incredible future ahead. 🚀🤖 #HumanoidRobots #AI #Robotics #FoundationRobotics #PhantomRobot #FutureOfWork

@luke_metro This is very accurate

Nvidia GTC 2026:

🚨⚡BREAKING NEWS: Foundation Robotics ran a pilot in Ukraine 🧵1/2 @sankaet says the first battlefield use case for @foundation_robo Phantom bot is simple, immediate & high impact: 📌Supply pickup 📌Battlefield logistics 📌Sending bots into risky tasks instead of soldiers

@mikekalilmfg Phantom mk1 can carry 40 kgs. Mk2 will be able to carry 80 kgs

Silicon Valley's latest humanoid is exceptionally strong and it's already doing real factory work. That's according to Noble Machines, a Sunnyvale-based startup launched in 2024 by former engineers from Apple, SpaceX, NASA, and Caltech. The startup says it's already shipped its first general purpose robot workers to an unnamed Fortune Global 500 customer. The headless humanoid robot can reportedly learn new industrial tasks in just a few hours by watching demonstrations or through natural language instructions. Standing 170 cm (5 feet 7 inches), it boasts a reported 34 degrees of freedom (DoF). The robot can lift and carry up to 27 kg (60 lbs), which makes it one of today's strongest robots on two legs. Despite its strength, Noble Machines, formerly known as Under Control Robotics (UCR), says its humanoid consumes an average of just 200 watts per hour, which is about the same amount of energy required for a household television. It can operate for five to six hours per shift, powered by a single NVIDIA Jetson Orin edge computer. Noble has not yet publicly disclosed pricing.

@cornelius_ong @TheHumanoidHub 🤐

@sankaet @TheHumanoidHub But don't you think planetary gearboxes can achieve similar performance while being significantly cheaper? 🤔

Foundation Robotics engineers describe the Phantom humanoid’s actuators. Low backdriving torque, low inertia, and high-bandwidth actuation through cycloidal drive.

@TheHumanoidHub @cornelius_ong just saw this... they are cheaper and weight less since they require less components. Atleast that is true for our design

@cornelius_ong Do you think cycloidal drives are worth the extra weight and cost?

Science education has been a passion of mine for a while. So we'll post more deep dives on various aspects of our technology. We’ve done hands and now head youtu.be/nfB8ySnDezg?si…

Ever wondered whats inside a humanoid head and how its different to biological intelligence? @foundation_robo team did a deep dive. Full video in 🧵

Both Phantom and I went on @MorningsMaria to discuss @foundation_robo and our collaboration with DoW and Ukraine and why I think deploying humanoids in the military is the best way to ensure lasting peace in our lifetime

Science education has been a passion of mine for a while. So we'll post more deep dives on various aspects of our technology. Here's the first on humanoid hands youtu.be/4BvLqO9iNLo?si…

@shapathdas @foundation_robo happy to explore.. @padsmagt will be the person worth talking to. Full disclosure - we are pretty advanced on this front on our side as well. This has been our focus for last 18 months and for 6-7 years for Patrick

@sankaet Hey Sanket, at Unsupervized we are building Physical Reasoning Model, we would be happy to explore strategic collaboration with @foundation_robo

Ever wondered whats inside a humanoid hand? @foundation_robo team did a deep dive on humanoid hands. Full video in 🧵

this was my first time attending a video shoot with @corememory and it was just incredible to watch the team put it together a must see on how foundation builds its humanoids