正在加载视频...

视频加载失败

返回首页

How to learn Physical Intelligence's robot brain without working at Physical Intelligence: PI was in talks this year to raise $1B at an $11B valuation, and its brain, π0, is open source three of its co-founders teach the ideas inside it on YouTube for $0: Sergey Levine, Chelsea Finn...

0x_meden's profile picture

Martin

1,666 subscribers

15,288 次观看 • 18 天前 •via X (Twitter)

科学技术教育
LIVE

Anya Rossi• Live Now

Private livecam show

6 条评论

Анастасиа Ел W 的头像
Анастасиа Ел W18 天前

one of my favorite things about PI is that they've open-sourced their models. really bullish on what this team is building.

Lloyd Lei 的头像
Lloyd Lei17 天前

This is what a company that defeated by OpenAI with actual-pre training ability looks like—raising 1B just to teach useless classes.

Sriram Gopalan 的头像
Sriram Gopalan18 天前

Good list. One thing lectures won’t teach: fine-tune π0 or SmolVLA on ~50 of your own episodes. You’ll quickly see how frame drops and inconsistent gripper closes show up as jitter in the policy.

PrismaX 的头像
PrismaX16 天前

π0.5 being open source changes the baseline for what a new team can ship on day one. The interesting constraint now is not model access but how the model was trained: what tasks, what robot, what data format. That's where the work starts, not ends.

Jenray 的头像
Jenray17 天前

Open-sourcing the brain is one thing; reproducing the data + teleop loop is usually the bottleneck. For someone following the Levine / Finn lectures: what’s the smallest setup that actually gets a π0-style policy to move beyond demos?

wujinlong 的头像
wujinlong18 天前

good work

相关视频

We built a Spiking Neural Network (SNN) to fully control a physical robot arm (6-DOF), and it's running entirely on a Raspberry Pi (no HAT). In our approach, the SNN starts untrained and drives the servos directly, learning in real time through local learning rules (no backprop or gradient descent). The goal with Twitch-y (the robot + SNN brain) is to let it explore and build an understanding of its own body until it can touch the plate on the mat, while also validating the SNN architecture and some of the math. In the video: - Top right: the SNN architecture - Bottom right: the mini-brain spiking in real time while controlling the arm - Left: Twitch-y (Hugging Face SO-101) If you want to learn more about Artificial Brains, come visit us this Friday (13th), at Founders Inc 👉
0:29

We built a Spiking Neural Network (SNN) to fully control a physical robot arm (6-DOF), and it's running entirely on a Raspberry Pi (no HAT). In our approach, the SNN starts untrained and drives the servos directly, learning in real time through local learning rules (no backprop or gradient descent). The goal with Twitch-y (the robot + SNN brain) is to let it explore and build an understanding of its own body until it can touch the plate on the mat, while also validating the SNN architecture and some of the math. In the video: - Top right: the SNN architecture - Bottom right: the mini-brain spiking in real time while controlling the arm - Left: Twitch-y (Hugging Face SO-101) If you want to learn more about Artificial Brains, come visit us this Friday (13th), at Founders Inc 👉

Alexander A. Wolf

30,665 次观看 • 9 个月前

Stanford charges $70,000 a year to teach robotics. I took their free lectures, put them in the right order and turned them into one website you can use for $0: it takes you from "never wrote a line of Python" to training your own robot policy in 6 months: → 119 free lectures from 7 courses, in the order a Stanford student actually takes them → taught by Chelsea Finn (co-founder of Physical Intelligence), Marco Pavone (NVIDIA), Oussama Khatib and the AI lead of Tesla Optimus → 6 projects in MuJoCo that run on a normal laptop, so you don't need a GPU or a robot to start → a 2-minute self-check that tells you which month to start from if you already code → a progress tracker that remembers every lecture and task you finish → a career map built from 412 real humanoid job posts, and only 2 of them require a PhD I built this for you guys, so if it's useful, reply "robot" and I'll keep updating it every day with new lectures, notebooks and open jobs link in the first reply 👇
0:43

Stanford charges $70,000 a year to teach robotics. I took their free lectures, put them in the right order and turned them into one website you can use for $0: it takes you from "never wrote a line of Python" to training your own robot policy in 6 months: → 119 free lectures from 7 courses, in the order a Stanford student actually takes them → taught by Chelsea Finn (co-founder of Physical Intelligence), Marco Pavone (NVIDIA), Oussama Khatib and the AI lead of Tesla Optimus → 6 projects in MuJoCo that run on a normal laptop, so you don't need a GPU or a robot to start → a 2-minute self-check that tells you which month to start from if you already code → a progress tracker that remembers every lecture and task you finish → a career map built from 412 real humanoid job posts, and only 2 of them require a PhD I built this for you guys, so if it's useful, reply "robot" and I'll keep updating it every day with new lectures, notebooks and open jobs link in the first reply 👇

Martin

11,508 次观看 • 17 天前

THIS ROBOT HAS A HUMAN FACE, AN OPEN TORSO FULL OF WIRES, AND SHE JUST TOOK A GLASS OF LEMON WATER SO NATURALLY THAT 200 PEOPLE STOPPED BREATHING! At first, most of the crowd thought it was a model in a costume. Then a team member handed her a glass of water with lemon, and she took it with both hands, adjusted her grip and looked at it like a person would. That is a much harder problem than it looks. The robot has to see the glass, judge its weight, control the force of each finger and avoid crushing or dropping it, all while a crowd is moving around her and phones are pointed at her face. Walking on stage was the old benchmark. Humans do small things like this thousands of times a day without thinking. A robot has to turn every one of them into sensors, motors and software. The strangest part is the contrast. The face and the movement say “human”, but the glowing mechanics in her torso and the metal in her leg say “machine”. Your brain keeps switching between the two and cannot settle on one. Every year the gap between “obviously a robot” and “wait, is that a person?” gets smaller.
0:26

THIS ROBOT HAS A HUMAN FACE, AN OPEN TORSO FULL OF WIRES, AND SHE JUST TOOK A GLASS OF LEMON WATER SO NATURALLY THAT 200 PEOPLE STOPPED BREATHING! At first, most of the crowd thought it was a model in a costume. Then a team member handed her a glass of water with lemon, and she took it with both hands, adjusted her grip and looked at it like a person would. That is a much harder problem than it looks. The robot has to see the glass, judge its weight, control the force of each finger and avoid crushing or dropping it, all while a crowd is moving around her and phones are pointed at her face. Walking on stage was the old benchmark. Humans do small things like this thousands of times a day without thinking. A robot has to turn every one of them into sensors, motors and software. The strangest part is the contrast. The face and the movement say “human”, but the glowing mechanics in her torso and the metal in her leg say “machine”. Your brain keeps switching between the two and cannot settle on one. Every year the gap between “obviously a robot” and “wait, is that a person?” gets smaller.

ssht

25,868 次观看 • 23 天前

This work makes a humanoid robot do simple parkour moves by looking with a depth camera and choosing the right move on the fly. The big deal is that it turns lots of small human moves into long, real-time robot behavior, without hand-coding every transition or retraining for each new course. A humanoid robot is usually good at steady walking, but it often fails when it has to do fast moves like jumping up, vaulting, or rolling, and then keep going to the next obstacle. The hard part is that you cannot easily collect training data for every possible obstacle shape, distance, and mistake, so robots end up learning a few moves that only work in a narrow setup. This work starts from short clips of real human parkour moves, like stepping over, vaulting, climbing, and rolling. It uses motion matching, which is basically a smart “pick the next clip that fits best right now” search, to stitch those short clips into a long, smooth plan that looks like a human doing a whole course. Then it trains a controller with reinforcement learning (RL), which means the robot learns by trial and error to copy that plan while staying balanced and not falling. After training separate expert controllers for different moves, it compresses them into 1 controller that uses only onboard depth sensing and a simple “go this fast in this direction” command. In real tests on a Unitree G1 humanoid, it can clear multiple obstacles in a row, adapt when obstacles get moved, and climb a wall up to 1.25m.
0:05

This work makes a humanoid robot do simple parkour moves by looking with a depth camera and choosing the right move on the fly. The big deal is that it turns lots of small human moves into long, real-time robot behavior, without hand-coding every transition or retraining for each new course. A humanoid robot is usually good at steady walking, but it often fails when it has to do fast moves like jumping up, vaulting, or rolling, and then keep going to the next obstacle. The hard part is that you cannot easily collect training data for every possible obstacle shape, distance, and mistake, so robots end up learning a few moves that only work in a narrow setup. This work starts from short clips of real human parkour moves, like stepping over, vaulting, climbing, and rolling. It uses motion matching, which is basically a smart “pick the next clip that fits best right now” search, to stitch those short clips into a long, smooth plan that looks like a human doing a whole course. Then it trains a controller with reinforcement learning (RL), which means the robot learns by trial and error to copy that plan while staying balanced and not falling. After training separate expert controllers for different moves, it compresses them into 1 controller that uses only onboard depth sensing and a simple “go this fast in this direction” command. In real tests on a Unitree G1 humanoid, it can clear multiple obstacles in a row, adapt when obstacles get moved, and climb a wall up to 1.25m.

Rohan Paul

37,121 次观看 • 8 个月前

In order for robots to be deployed in the real world, performing tasks of real value, they must be reliable. Unfortunately, even more, most robotic demos work maybe 70-80% of the time at best. The way to get better reliability is to do real-world reinforcement learning: having the robot teach itself how to perform the task up to a high level of success. The key to doing this is to start with a core of expert human data, use that to train a policy then iteratively improve it, until finally finishing with on-policy reinforcement learning. Kun Lei talks through a unified framework for imitation and reinforcement learning based on PPO, which enables this improvement process. In this episode, Kun Lei explains the theory behind his reinforcement learning method and how it allowed his robot to run in a shopping mall juicing oranges for seven hours at a time, among experiments on a wide variety of tasks and embodiments. Watch episode 58 of RoboPapers now, hosted by Michael Cho - Rbt/Acc and Chris Paxton!
1:11:51

In order for robots to be deployed in the real world, performing tasks of real value, they must be reliable. Unfortunately, even more, most robotic demos work maybe 70-80% of the time at best. The way to get better reliability is to do real-world reinforcement learning: having the robot teach itself how to perform the task up to a high level of success. The key to doing this is to start with a core of expert human data, use that to train a policy then iteratively improve it, until finally finishing with on-policy reinforcement learning. Kun Lei talks through a unified framework for imitation and reinforcement learning based on PPO, which enables this improvement process. In this episode, Kun Lei explains the theory behind his reinforcement learning method and how it allowed his robot to run in a shopping mall juicing oranges for seven hours at a time, among experiments on a wide variety of tasks and embodiments. Watch episode 58 of RoboPapers now, hosted by Michael Cho - Rbt/Acc and Chris Paxton!

RoboPapers

18,813 次观看 • 10 个月前

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:
0:29

Tired of teleoperation? One human video → 1,000s of robot demos. (📍GitHub ) Scaling Robot Data Without Dynamics Simulation or Robot Hardware Real2Render2Real (R2R2R) is a new way to scale robot data without physics simulation or hardware. You take a phone scan + a single monocular human demo. It tracks the motion, renders photorealistic scenes, and generates diverse, robot-agnostic trajectories ready for training. > No teleop, no sim, no robot, just a phone and a video > Train VLA models and diffusion policies directly on the output > Supports multiple robot embodiments with kinematic consistency > 1000s of demos in 1/27 the time of real-world collection Thank you, Max Fu, for sharing!! Project: Paper: Code coming soon: It shows that with the right pipeline, you can scale robot learning data without touching a robot. One of the most interesting directions in scalable robotics today. —— Weekly robotics and AI insights. Subscribe free:

Ilir Aliu

42,864 次观看 • 9 个月前

Enough with LLMs - exciting things are happening in the world of atoms. This is Stanford ALOHA, a low-cost and agile robot platform. The whole system is open-source (!!): hardware design, CAD models for 3D printing, simulator, and training code. Time to graft a physical arm onto GPTs 🦾 Led by my friend Tony Zhao at Stanford, and advised by Chelsea Finn Sergey Levine Vikash Kumar ✈️IROS2026. Project page: If you don't want to 3D print your own components, you can buy the setup at
1:46

Enough with LLMs - exciting things are happening in the world of atoms. This is Stanford ALOHA, a low-cost and agile robot platform. The whole system is open-source (!!): hardware design, CAD models for 3D printing, simulator, and training code. Time to graft a physical arm onto GPTs 🦾 Led by my friend Tony Zhao at Stanford, and advised by Chelsea Finn Sergey Levine Vikash Kumar ✈️IROS2026. Project page: If you don't want to 3D print your own components, you can buy the setup at

Jim Fan

819,054 次观看 • 3 年前

A Switzerland-based startup Flexion has created a robotic brain that helps the Unitree G1 move smoothly and work on its own. It uses reinforcement learning, where the robot trains in simulations to learn walking, balancing, and picking objects. In tests, it cleaned a space by finding and placing items in a basket without human help.
0:47

A Switzerland-based startup Flexion has created a robotic brain that helps the Unitree G1 move smoothly and work on its own. It uses reinforcement learning, where the robot trains in simulations to learn walking, balancing, and picking objects. In tests, it cleaned a space by finding and placing items in a basket without human help.

Space and Technology

17,557 次观看 • 7 个月前

🇺🇸 TESLA’S OPTIMUS ROBOT JUST STOLE THE SHOW IN MIAMI It didn’t give a speech, but it didn’t need to. Just by waving and handing out water, Optimus got the kind of attention you'd expect a real, functioning humanoid robot to get. At Tesla’s “Future of Autonomy” event, Optimus wasn’t just for show. It moved on its own, interacted with people, and looked one step closer to something out of a sci-fi movie. Tesla says Optimus is trained using the same tech behind Autopilot, feeding on millions of real-world videos to learn how to operate in messy human environments. Source: Tesla, Erick Brito
0:41

🇺🇸 TESLA’S OPTIMUS ROBOT JUST STOLE THE SHOW IN MIAMI It didn’t give a speech, but it didn’t need to. Just by waving and handing out water, Optimus got the kind of attention you'd expect a real, functioning humanoid robot to get. At Tesla’s “Future of Autonomy” event, Optimus wasn’t just for show. It moved on its own, interacted with people, and looked one step closer to something out of a sci-fi movie. Tesla says Optimus is trained using the same tech behind Autopilot, feeding on millions of real-world videos to learn how to operate in messy human environments. Source: Tesla, Erick Brito

Mario Nawfal

145,228 次观看 • 11 个月前

4. $AMBA - Ambarella The brain that rides along inside the machine. Chips small enough to run AI on the robot itself, so it decides in real time instead of waiting on a data center. NXP is reportedly in talks to buy them for ~$3.3 BILLION.
0:45

4. $AMBA - Ambarella The brain that rides along inside the machine. Chips small enough to run AI on the robot itself, so it decides in real time instead of waiting on a data center. NXP is reportedly in talks to buy them for ~$3.3 BILLION.

Frank Curzio

159,633 次观看 • 2 个月前

you don’t need a robot to see physical ai in action, you just need a browser. this week I finally tried something I've wanted to do for a while: design an SO-101 scene in three.js, connect it to a real leader arm, collect teleoperation data, train an ACT model, then port it to run in the browser. the result is the first ever demo of ACT that runs locally on the web. try it yourself now at
0:35

you don’t need a robot to see physical ai in action, you just need a browser. this week I finally tried something I've wanted to do for a while: design an SO-101 scene in three.js, connect it to a real leader arm, collect teleoperation data, train an ACT model, then port it to run in the browser. the result is the first ever demo of ACT that runs locally on the web. try it yourself now at

Binh

122,825 次观看 • 4 个月前

Force is arguably the most overlooked ingredient in modern robot learning. Introducing FACTR 2: it turns *any* commodity robot into a force-aware system with no force sensors required. Train a tiny force network in <1min with <10mins of data and drop it into any existing teleop pipelines: ✅ Free force sensing for both the robot and the operator arm ✅ Makes demos higher-quality → fewer of them needed. ✅ A new force-aware learning algorithm (FIRST) uses those recovered forces to figure out which parts of a demo actually matter, making learning data-efficient. ✅ Strong performance on complex tasks with fewer demos and even no pretraining! More details below.
0:40

Force is arguably the most overlooked ingredient in modern robot learning. Introducing FACTR 2: it turns *any* commodity robot into a force-aware system with no force sensors required. Train a tiny force network in <1min with <10mins of data and drop it into any existing teleop pipelines: ✅ Free force sensing for both the robot and the operator arm ✅ Makes demos higher-quality → fewer of them needed. ✅ A new force-aware learning algorithm (FIRST) uses those recovered forces to figure out which parts of a demo actually matter, making learning data-efficient. ✅ Strong performance on complex tasks with fewer demos and even no pretraining! More details below.

Deepak Pathak

41,203 次观看 • 5 个月前

$BRAIN is live. the open-source fly-brain bridge has a ticker now, and the ticker is becoming the input for a physical robot. CA: 9LuwgFQemAoV9rgVBBtwbRSxBmamcRRbysEK8yL2pump what’s already done: the neural bridge is working. camera input becomes activity across eight virtual neural populations, and that activity becomes left and right motor commands IMU data feeds movement back into the model. smoothing, speed limits and a 500 ms watchdog are already built in the synthetic demo runs locally today. the ESP32 scaffold is ready for hardware integration the entire project is open source and MIT licensed github: what’s next, in order: the on-chain listener goes live. every $BRAIN transaction becomes a stimulus sent directly into the neural model then the physical robot connects. the wallet address determines which neurons activate, the amount determines the strength of the impulse, and the brain converts that reaction into movement then the livestream. every transaction, neural impulse and physical response visible in real time then user commands. spend $BRAIN to trigger a stronger reflex, request movement, temporarily control the robot, name a neuron or place your name on the stream the loop, plain: a transaction enters the neurons react the robot moves a Proof of Reflex is created with the transaction hash, neural activity map and video of the physical response the current brain is a small fly-inspired simulation. the bridge and local demo work today. the blockchain connection, physical robot and livestream are being built now the internet becomes its sensory organ. the blockchain becomes its nervous system. $BRAIN makes the body move.
0:20

$BRAIN is live. the open-source fly-brain bridge has a ticker now, and the ticker is becoming the input for a physical robot. CA: 9LuwgFQemAoV9rgVBBtwbRSxBmamcRRbysEK8yL2pump what’s already done: the neural bridge is working. camera input becomes activity across eight virtual neural populations, and that activity becomes left and right motor commands IMU data feeds movement back into the model. smoothing, speed limits and a 500 ms watchdog are already built in the synthetic demo runs locally today. the ESP32 scaffold is ready for hardware integration the entire project is open source and MIT licensed github: what’s next, in order: the on-chain listener goes live. every $BRAIN transaction becomes a stimulus sent directly into the neural model then the physical robot connects. the wallet address determines which neurons activate, the amount determines the strength of the impulse, and the brain converts that reaction into movement then the livestream. every transaction, neural impulse and physical response visible in real time then user commands. spend $BRAIN to trigger a stronger reflex, request movement, temporarily control the robot, name a neuron or place your name on the stream the loop, plain: a transaction enters the neurons react the robot moves a Proof of Reflex is created with the transaction hash, neural activity map and video of the physical response the current brain is a small fly-inspired simulation. the bridge and local demo work today. the blockchain connection, physical robot and livestream are being built now the internet becomes its sensory organ. the blockchain becomes its nervous system. $BRAIN makes the body move.

Frank

347,724 次观看 • 2 个月前

A toy robot. A home assistant. A humanoid on a factory floor. They all need the same thing: a brain. QVAC is that brain, and it runs on the robot itself. No cloud to phone, no latency, nothing leaving your home. The intelligence is on the machine, so it keeps thinking even with the internet off. The robots are coming. This is what wakes them up:
0:37

A toy robot. A home assistant. A humanoid on a factory floor. They all need the same thing: a brain. QVAC is that brain, and it runs on the robot itself. No cloud to phone, no latency, nothing leaving your home. The intelligence is on the machine, so it keeps thinking even with the internet off. The robots are coming. This is what wakes them up:

QVAC

5,250,097 次观看 • 4 个月前

Ten million people have watched an MIT professor teach a course whose first lecture is literally titled "What is a Derivative?" Almost none have written down the two-line answer. He filmed the lecture once in the fall of 2007 and it has been on YouTube ever since. Math tutors charge $200 an hour to teach a diluted version of what he covered in 50 minutes for free. His name is David Jerison. He is a professor of mathematics at MIT and the instructor of 18.01 Single Variable Calculus, one of the most-watched math courses in the history of the internet. The 50-minute clip in this video is Lecture 1, filmed at MIT in the fall of 2007. Jerison is deriving the definition of a derivative from a single tangent line. The whole framework fits on a napkin. A derivative is just how much y changes when x moves a tiny bit. Draw a tangent line to any curve at any point. The slope of that line is the derivative. Memorize one formula, the power rule, and you can differentiate every polynomial on earth in your head. Chain that with a handful of exceptions and you can differentiate almost every function humanity has ever written down. That single set of rules is what every neural network runs on gradient descent, what every rocket landing at SpaceX solves in real time, and what every options desk at Goldman Sachs is running behind every quote you see on the screen. "In mathematics you don't understand things. You just get used to them." That is John von Neumann, the mathematician who helped design the atomic bomb and invent the modern computer. Jerison returns to the same idea in every lecture. Almost no student giving up on calculus has heard von Neumann say it out loud. Every quant fund on Wall Street pays entry-level analysts $250,000 to know the same power rule Jerison derives on the board. Every AI bootcamp charges tens of thousands to teach a diluted version of the same equation on a laptop. The lecture is free on MIT OpenCourseWare. The textbook is under sixty dollars. Almost none of the millions who watched have ever taken the power rule and applied it to their own numbers on their own paper. The math is free. The willingness to actually take one derivative before your next model, trade, or engineering trade-off is the entire edge.
50:48

Ten million people have watched an MIT professor teach a course whose first lecture is literally titled "What is a Derivative?" Almost none have written down the two-line answer. He filmed the lecture once in the fall of 2007 and it has been on YouTube ever since. Math tutors charge $200 an hour to teach a diluted version of what he covered in 50 minutes for free. His name is David Jerison. He is a professor of mathematics at MIT and the instructor of 18.01 Single Variable Calculus, one of the most-watched math courses in the history of the internet. The 50-minute clip in this video is Lecture 1, filmed at MIT in the fall of 2007. Jerison is deriving the definition of a derivative from a single tangent line. The whole framework fits on a napkin. A derivative is just how much y changes when x moves a tiny bit. Draw a tangent line to any curve at any point. The slope of that line is the derivative. Memorize one formula, the power rule, and you can differentiate every polynomial on earth in your head. Chain that with a handful of exceptions and you can differentiate almost every function humanity has ever written down. That single set of rules is what every neural network runs on gradient descent, what every rocket landing at SpaceX solves in real time, and what every options desk at Goldman Sachs is running behind every quote you see on the screen. "In mathematics you don't understand things. You just get used to them." That is John von Neumann, the mathematician who helped design the atomic bomb and invent the modern computer. Jerison returns to the same idea in every lecture. Almost no student giving up on calculus has heard von Neumann say it out loud. Every quant fund on Wall Street pays entry-level analysts $250,000 to know the same power rule Jerison derives on the board. Every AI bootcamp charges tens of thousands to teach a diluted version of the same equation on a laptop. The lecture is free on MIT OpenCourseWare. The textbook is under sixty dollars. Almost none of the millions who watched have ever taken the power rule and applied it to their own numbers on their own paper. The math is free. The willingness to actually take one derivative before your next model, trade, or engineering trade-off is the entire edge.

Lumen

31,497 次观看 • 1 个月前

The future of sports is evolving in China. A robot just played tennis against a human at the World Humanoid Robot Games in Beijing with zero human control, tracking the ball and returning shots entirely on its own. How long before robots have their own Olympics?
1:19

The future of sports is evolving in China. A robot just played tennis against a human at the World Humanoid Robot Games in Beijing with zero human control, tracking the ball and returning shots entirely on its own. How long before robots have their own Olympics?

Coin Bureau

57,188 次观看 • 2 个月前

Gemini + π0 = actually useful robots! (Similar to what Physical Intelligence did with "Hi Robot") I can now verbally tell the robot that I'm building a red Lego wall or wooden tower, and it will infer the next steps by itself and pass me the necessary pieces, tools, or materials, ha! You can also just ask it to bring you things! The pipeline works as follows: - OpenAI Whisper (local) → speech to text - Gemini → makes sense of user requests, converts to robot tasks, bounding boxes, grasping points, etc. (System 2 thinking FTW!) - π0 → robotic actions The π0 was finetuned just for pick-and-place Lego bricks only, and it generalizes beautifully to all kinds of tasks. However, there's lots of room for improvement when it comes to grasping & accuracy. Things that could help: - Conditioning on grasping points - Better data collection (I'm not that great at teleop) - Lots more synthetic data and simulations
1:04

Gemini + π0 = actually useful robots! (Similar to what Physical Intelligence did with "Hi Robot") I can now verbally tell the robot that I'm building a red Lego wall or wooden tower, and it will infer the next steps by itself and pass me the necessary pieces, tools, or materials, ha! You can also just ask it to bring you things! The pipeline works as follows: - OpenAI Whisper (local) → speech to text - Gemini → makes sense of user requests, converts to robot tasks, bounding boxes, grasping points, etc. (System 2 thinking FTW!) - π0 → robotic actions The π0 was finetuned just for pick-and-place Lego bricks only, and it generalizes beautifully to all kinds of tasks. However, there's lots of room for improvement when it comes to grasping & accuracy. Things that could help: - Conditioning on grasping points - Better data collection (I'm not that great at teleop) - Lots more synthetic data and simulations

Shreyas Gite

25,505 次观看 • 1 年前

Ten million people have watched a Stanford lecturer accidentally destroy the entire public speaking coaching industry. He filmed the lecture once in October 2014 and it has been on YouTube ever since. Speaking coaches charge two thousand dollars a session to teach a third of what he covered in one hour for free. His name is Matt Abrahams. He is a lecturer at the Stanford Graduate School of Business and the coach who has trained thousands of Silicon Valley founders on how to answer questions live. Every year at Stanford's Alumni Weekend, he gives the same lecture called "Think Fast, Talk Smart." His entire framework fits on a napkin. Anxiety is not a bug. Greet it. Do not perform. Have a conversation. Never open with an apology. Use What, So What, Now What when your brain freezes. Problem, Solution, Benefit for a pitch. Never end with "any questions?" That last rule alone has probably cost the pitch coaching industry a hundred million dollars. "Ninety percent of the words that will decide your career are not written down in advance." That is the actual opening line of the lecture. Abrahams built the entire Stanford elective on the sentence. Founders spend $50,000 on pitch coaches and freeze the moment the investor asks a follow-up question. Engineers walk into performance reviews having rehearsed every accomplishment and blank on the one behavioral question that decides the promotion. The lecture is free on Stanford's YouTube channel. Abrahams's book "Think Faster, Talk Smarter" is under twenty dollars. Abrahams still teaches at Stanford. Almost none of the ten million viewers have used the three-step frame when put on the spot. The frame is free. The willingness to actually use it the next time someone puts you on the spot is the entire edge.
58:09

Ten million people have watched a Stanford lecturer accidentally destroy the entire public speaking coaching industry. He filmed the lecture once in October 2014 and it has been on YouTube ever since. Speaking coaches charge two thousand dollars a session to teach a third of what he covered in one hour for free. His name is Matt Abrahams. He is a lecturer at the Stanford Graduate School of Business and the coach who has trained thousands of Silicon Valley founders on how to answer questions live. Every year at Stanford's Alumni Weekend, he gives the same lecture called "Think Fast, Talk Smart." His entire framework fits on a napkin. Anxiety is not a bug. Greet it. Do not perform. Have a conversation. Never open with an apology. Use What, So What, Now What when your brain freezes. Problem, Solution, Benefit for a pitch. Never end with "any questions?" That last rule alone has probably cost the pitch coaching industry a hundred million dollars. "Ninety percent of the words that will decide your career are not written down in advance." That is the actual opening line of the lecture. Abrahams built the entire Stanford elective on the sentence. Founders spend $50,000 on pitch coaches and freeze the moment the investor asks a follow-up question. Engineers walk into performance reviews having rehearsed every accomplishment and blank on the one behavioral question that decides the promotion. The lecture is free on Stanford's YouTube channel. Abrahams's book "Think Faster, Talk Smarter" is under twenty dollars. Abrahams still teaches at Stanford. Almost none of the ten million viewers have used the three-step frame when put on the spot. The frame is free. The willingness to actually use it the next time someone puts you on the spot is the entire edge.

Lily

66,201 次观看 • 1 个月前

Ten million people have watched a Stanford lecturer accidentally destroy the entire public speaking coaching industry. He filmed the lecture once in October 2014 and it has been on YouTube ever since. Speaking coaches charge two thousand dollars a session to teach a third of what he covered in one hour for free. His name is Matt Abrahams. He is a lecturer at the Stanford Graduate School of Business and the coach who has trained thousands of Silicon Valley founders on how to answer questions live. Every year at Stanford's Alumni Weekend, he gives the same lecture called "Think Fast, Talk Smart." His entire framework fits on a napkin. Anxiety is not a bug. Greet it. Do not perform. Have a conversation. Never open with an apology. Use What, So What, Now What when your brain freezes. Problem, Solution, Benefit for a pitch. Never end with "any questions?" That last rule alone has probably cost the pitch coaching industry a hundred million dollars. "Ninety percent of the words that will decide your career are not written down in advance." That is the actual opening line of the lecture. Abrahams built the entire Stanford elective on the sentence. Founders spend $50,000 on pitch coaches and freeze the moment the investor asks a follow-up question. Engineers walk into performance reviews having rehearsed every accomplishment and blank on the one behavioral question that decides the promotion. The lecture is free on Stanford's YouTube channel. Abrahams's book "Think Faster, Talk Smarter" is under twenty dollars. Abrahams still teaches at Stanford. Almost none of the ten million viewers have used the three-step frame when put on the spot. The frame is free. The willingness to actually use it the next time someone puts you on the spot is the entire edge.
58:09

Ten million people have watched a Stanford lecturer accidentally destroy the entire public speaking coaching industry. He filmed the lecture once in October 2014 and it has been on YouTube ever since. Speaking coaches charge two thousand dollars a session to teach a third of what he covered in one hour for free. His name is Matt Abrahams. He is a lecturer at the Stanford Graduate School of Business and the coach who has trained thousands of Silicon Valley founders on how to answer questions live. Every year at Stanford's Alumni Weekend, he gives the same lecture called "Think Fast, Talk Smart." His entire framework fits on a napkin. Anxiety is not a bug. Greet it. Do not perform. Have a conversation. Never open with an apology. Use What, So What, Now What when your brain freezes. Problem, Solution, Benefit for a pitch. Never end with "any questions?" That last rule alone has probably cost the pitch coaching industry a hundred million dollars. "Ninety percent of the words that will decide your career are not written down in advance." That is the actual opening line of the lecture. Abrahams built the entire Stanford elective on the sentence. Founders spend $50,000 on pitch coaches and freeze the moment the investor asks a follow-up question. Engineers walk into performance reviews having rehearsed every accomplishment and blank on the one behavioral question that decides the promotion. The lecture is free on Stanford's YouTube channel. Abrahams's book "Think Faster, Talk Smarter" is under twenty dollars. Abrahams still teaches at Stanford. Almost none of the ten million viewers have used the three-step frame when put on the spot. The frame is free. The willingness to actually use it the next time someone puts you on the spot is the entire edge.

Lumen

139,166 次观看 • 1 个月前