Harrison Kinsley's banner

Harrison Kinsley

@Sentdex • 104,520 subscribers

gpus and tractors. Director of AI and Engineering @ https://t.co/H4St8dd1ip Neural networks from Scratch book: https://t.co/hyMkWyUP7R https://t.co/8WGZRkUGsn

Shorts

This is incredible

This is incredible

2,846,098 просмотров

testing robot policies has never been so much fun

testing robot policies has never been so much fun

57,986 просмотров

on a scale of impressive to theatrics, where do you find the ceo of engine getting kicked by his robot to land Brett Adcock ?

on a scale of impressive to theatrics, where do you find the ceo of engine getting kicked by his robot to land Brett Adcock ?

92,560 просмотров

Playing on the new Jetson Thor trying to think in terms of having gobs of memory, but low mem bandwidth Moondream2 VLM is ~2 FPS per full loop of everything But we have 128GB of memory. So run 15 VLM servers (~76GB) & get 30 FPS w/ ~100ms latency for the feed very comfy.

Playing on the new Jetson Thor trying to think in terms of having gobs of memory, but low mem bandwidth Moondream2 VLM is ~2 FPS per full loop of everything But we have 128GB of memory. So run 15 VLM servers (~76GB) & get 30 FPS w/ ~100ms latency for the feed very comfy.

35,687 просмотров

Videos

Anya Rossi

sweetdream.ai

SweetDream.ai•Sponsored•Livecam

Watch Anya Live

Anya is streaming live right now! Join her private show and enjoy exclusive content.

Exclusive private shows

1.2k viewers online

Private Show

Join now for exclusive access

Free preview available • Premium content

People often ask how did the Unitree robots get so good all of a sudden. It wasn't all of a sudden, and it's because they ship their hardware and open source their SDKs. Arguably these robots are nearly useless out of the box, but you have full dev control of them. Because of that, the hardware has become a very popular R&D platform with an ecosystem around it and the Unitree G1 is undoubtedly an order of magnitude better than it could ever be at this point if Unitree was instead just doing quiet internal dev of both the hardware and software. Too many hardware companies for really cool products that seek to be community-driven (robots, AR glasses...etc) desire to make a profitable walled garden and their greed just ends up walling out developers and their product gets outpaced by the G1s of the world.

People often ask how did the Unitree robots get so good all of a sudden. It wasn't all of a sudden, and it's because they ship their hardware and open source their SDKs. Arguably these robots are nearly useless out of the box, but you have full dev control of them. Because of that, the hardware has become a very popular R&D platform with an ecosystem around it and the Unitree G1 is undoubtedly an order of magnitude better than it could ever be at this point if Unitree was instead just doing quiet internal dev of both the hardware and software. Too many hardware companies for really cool products that seek to be community-driven (robots, AR glasses...etc) desire to make a profitable walled garden and their greed just ends up walling out developers and their product gets outpaced by the G1s of the world.

Harrison Kinsley

815,526 просмотров • 8 месяцев назад

I need a picker upper pupper in my twin toddler life.

I need a picker upper pupper in my twin toddler life.

Harrison Kinsley

13,751 просмотров • 25 дней назад

Tesla a bit too keen to prove it can do offroad. Video from a family member's Tesla in FSD v12.6.4 ping-ponging, hopping a driveway, and ending up in a ditch. No one injured except the car, but damn. FSD is truly amazing until it isn't.

Tesla a bit too keen to prove it can do offroad. Video from a family member's Tesla in FSD v12.6.4 ping-ponging, hopping a driveway, and ending up in a ditch. No one injured except the car, but damn. FSD is truly amazing until it isn't.

Harrison Kinsley

104,672 просмотров • 1 год назад

just cooked up a new sprinter policy, do we attempt sim2real?

just cooked up a new sprinter policy, do we attempt sim2real?

Harrison Kinsley

57,912 просмотров • 7 месяцев назад

picking something up off the floor w/ a humanoid is more challenging than a backflip

picking something up off the floor w/ a humanoid is more challenging than a backflip

Harrison Kinsley

25,485 просмотров • 3 месяцев назад

In a world of PPO everything for reinforcement learning, I've been tinkering with SAC for training a quadruped gait. This gait is trained purely on CPU (training on one of the Dell GB10s) on a single environment. Training any particular run is obviously slower than PPO on an RTX Pro 6000 with 8092 envs, if you already know the exact hyperparams/rwd function for your PPO algo... but, if we're honest with ourselves, then we know we usually spend days tuning our PPO algo and fighting it to do what we want. In contrast, SAC has kind of been a breath of fresh air, very amenable to changing the reward function to tune behavior. So far, my first attempts to tune things have consistently just worked immediately rather than 15 different variations of reward hacking only to find previous tuned behaviors got lost in the process. There is also FastSAC, which I've not yet tried, but can speed things up potentially and introduce scale back into the equation. My main painpoint in getting SAC to work for gait was actually getting it to learn to step. It seems as though SAC is not as good as PPO at significant exploration on its own. I ended up starting with a sinusoidal gait (basically just a rule to make legs swing) as training wheels then blended it out through training as phase 1, then began working on smoothing things out after this. I think if we look at end to end dev time rather than any particular run that finally managed to work, SAC may actually be the "faster" algorithm to train. Quadruped gaits are inherently easier than bipedal and maybe there are areas where SAC falls short, but I'll definitely be spending more time with SAC.

In a world of PPO everything for reinforcement learning, I've been tinkering with SAC for training a quadruped gait. This gait is trained purely on CPU (training on one of the Dell GB10s) on a single environment. Training any particular run is obviously slower than PPO on an RTX Pro 6000 with 8092 envs, if you already know the exact hyperparams/rwd function for your PPO algo... but, if we're honest with ourselves, then we know we usually spend days tuning our PPO algo and fighting it to do what we want. In contrast, SAC has kind of been a breath of fresh air, very amenable to changing the reward function to tune behavior. So far, my first attempts to tune things have consistently just worked immediately rather than 15 different variations of reward hacking only to find previous tuned behaviors got lost in the process. There is also FastSAC, which I've not yet tried, but can speed things up potentially and introduce scale back into the equation. My main painpoint in getting SAC to work for gait was actually getting it to learn to step. It seems as though SAC is not as good as PPO at significant exploration on its own. I ended up starting with a sinusoidal gait (basically just a rule to make legs swing) as training wheels then blended it out through training as phase 1, then began working on smoothing things out after this. I think if we look at end to end dev time rather than any particular run that finally managed to work, SAC may actually be the "faster" algorithm to train. Quadruped gaits are inherently easier than bipedal and maybe there are areas where SAC falls short, but I'll definitely be spending more time with SAC.

Harrison Kinsley

26,711 просмотров • 3 месяцев назад

This is a vertically integrated end to end deep neural network performing forward pass inference real-time, controlling individual actuator's torque output for bidpedal gait generation in adverse, GPS denied, envs. ok its standard PPO rl trained in mjlab, strapped to a tractor.

This is a vertically integrated end to end deep neural network performing forward pass inference real-time, controlling individual actuator's torque output for bidpedal gait generation in adverse, GPS denied, envs. ok its standard PPO rl trained in mjlab, strapped to a tractor.

Harrison Kinsley

34,118 просмотров • 5 месяцев назад

That's cute, you can do some bunny hopping? I'm jumping into another dimension, learning the meaning of everything and coming back.

That's cute, you can do some bunny hopping? I'm jumping into another dimension, learning the meaning of everything and coming back.

Harrison Kinsley

80,291 просмотров • 1 год назад

Ladies and gentlemen, we have our first successful sim2real transfer on Geoff the G1!

Ladies and gentlemen, we have our first successful sim2real transfer on Geoff the G1!

Harrison Kinsley

28,619 просмотров • 6 месяцев назад

after many failed attempts to make a single model that walks and crouches with all the attributes I wanted, I ended up just splitting into 2 models with basic control logic to swap between them. Could just keep adding "skills" as models and tweak those specific models as needed

after many failed attempts to make a single model that walks and crouches with all the attributes I wanted, I ended up just splitting into 2 models with basic control logic to swap between them. Could just keep adding "skills" as models and tweak those specific models as needed

Harrison Kinsley

19,212 просмотров • 4 месяцев назад

alright, finally we're almost ready to ship our new policy to the world

alright, finally we're almost ready to ship our new policy to the world

Harrison Kinsley

20,756 просмотров • 7 месяцев назад

further reward crafting for hand-centric (tm) PPO model. Closing in on what I want. need to slow down max velocities and I want more stepping/stride + less tip toe on left foot ideally. almost cant believe this is actually just RL!

further reward crafting for hand-centric (tm) PPO model. Closing in on what I want. need to slow down max velocities and I want more stepping/stride + less tip toe on left foot ideally. almost cant believe this is actually just RL!

Harrison Kinsley

15,971 просмотров • 5 месяцев назад

"Random Person" interview of Lisa Su, CEO of AMD and sponsor of Ferrari at the time, on grid at a Formula 1 race (2018)

"Random Person" interview of Lisa Su, CEO of AMD and sponsor of Ferrari at the time, on grid at a Formula 1 race (2018)

Harrison Kinsley

30,482 просмотров • 2 лет назад

The next long-form installment of development with the Unitree G1 Humanoid using codex and o3, covering LiDAR/SLAM, more on manual navigation, route planning, and my general experience up to this point with the G1!

The next long-form installment of development with the Unitree G1 Humanoid using codex and o3, covering LiDAR/SLAM, more on manual navigation, route planning, and my general experience up to this point with the G1!

Harrison Kinsley

14,297 просмотров • 1 год назад

We be unboxing the Unitree G1 Edu Ultimate humanoid and getting our feet wet with programming it. There simply is no better unboxing video.

We be unboxing the Unitree G1 Edu Ultimate humanoid and getting our feet wet with programming it. There simply is no better unboxing video.

Harrison Kinsley

13,080 просмотров • 1 год назад

Больше нет контента для загрузки