Загрузка видео...

Не удалось загрузить видео

На главную

right now my reinforcement learning model is basically doing this and i came up with a solution for it all by myself (by stealing it from pufferlib) that not only solves it, but also helps keep training stable and fast

yacineMTB's profile picture

kache

201,365 subscribers

60,094 просмотров • 10 месяцев назад •via X (Twitter)

Наука и технологииОбразованиеИскусство
LIVE

Anya Rossi• Live Now

Private livecam show

Комментарии: 0

Нет доступных комментариев

Здесь появятся комментарии из оригинального поста

Похожие видео

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 просмотров • 5 месяцев назад

🐶: changbin wrote this a year ago as a birthday present. i recorded it last week by myself and i worked on the tune myself and i mixed it myself and i sent it to hyung. hyung liked it but it's not the final version yet so i'll polish it a bit more and release it for stay~
0:18

🐶: changbin wrote this a year ago as a birthday present. i recorded it last week by myself and i worked on the tune myself and i mixed it myself and i sent it to hyung. hyung liked it but it's not the final version yet so i'll polish it a bit more and release it for stay~

lena ✰

62,178 просмотров • 8 месяцев назад

Finally, I now own a laptop! If someone had told me in January that I'll own a laptop by now, I wouldn't believe it. But God honestly came through. Might not be the best out there but I'm heavy on starting small and growing gradually.💜 Keep showing up with that phone because it'll definitely be worth it at the end.
0:12

Finally, I now own a laptop! If someone had told me in January that I'll own a laptop by now, I wouldn't believe it. But God honestly came through. Might not be the best out there but I'm heavy on starting small and growing gradually.💜 Keep showing up with that phone because it'll definitely be worth it at the end.

𝐃𝐄𝐂𝐄𝐍𝐂𝐘💜

210,420 просмотров • 1 год назад

I came up with this silly idea after an art block and rotting from boredom. I hope someone likes it. I’m calling it Pointer Pet . It’s basically a pet that hovers next to a vtuber model and is a mouse tracker where the pet bites the cursor when you click. #live2dwip #vtuberassets #live2dwip
0:12

I came up with this silly idea after an art block and rotting from boredom. I hope someone likes it. I’m calling it Pointer Pet . It’s basically a pet that hovers next to a vtuber model and is a mouse tracker where the pet bites the cursor when you click. #live2dwip #vtuberassets #live2dwip

Chisa ♡ Live2D Artist

21,815 просмотров • 4 месяцев назад

I animated them all with my finger using flipaclip 32 gb ipad mini storage. All of them are unsuccessful. But all of my failure helps me improve myself until now. this app is a junk now 👎. I don’t use it anymore.
0:10

I animated them all with my finger using flipaclip 32 gb ipad mini storage. All of them are unsuccessful. But all of my failure helps me improve myself until now. this app is a junk now 👎. I don’t use it anymore.

FizzyNCasey

54,784 просмотров • 1 год назад

BOYNEXTDOOR's Jaehyun at Music Bank mentions doing a dance challenge solo since he couldn’t do one with #Jungkook 😭 🐶 “since I can’t film with Jungkook sunbaenim, I’ll just do it alone, by myself…! Ah, I practiced hard today, but I got nervous so it didn’t go well. I’ll film it separately by myself and upload it. Please watch it a lot, and I’ll keep working hard. Thank you.” Cr. phille1204
0:19

BOYNEXTDOOR's Jaehyun at Music Bank mentions doing a dance challenge solo since he couldn’t do one with #Jungkook 😭 🐶 “since I can’t film with Jungkook sunbaenim, I’ll just do it alone, by myself…! Ah, I practiced hard today, but I got nervous so it didn’t go well. I’ll film it separately by myself and upload it. Please watch it a lot, and I’ll keep working hard. Thank you.” Cr. phille1204

𝗥𝗔𝗣𝗣𝗘𝗥 𝗝𝗞

81,678 просмотров • 9 месяцев назад

My biggest passion right now is coding agents that figure out the storyboard and prompts from a script, as well as the ingredients (characters, locations, props). And it does all of it with great camera control and direction. It's not perfect yet, but it gets better every day!
0:15

My biggest passion right now is coding agents that figure out the storyboard and prompts from a script, as well as the ingredients (characters, locations, props). And it does all of it with great camera control and direction. It's not perfect yet, but it gets better every day!

Alex Patrascu

19,492 просмотров • 1 месяц назад

i went to a party by myself last night. got on the stage, danced, threw my leg up and hit a split. the crowd enjoyed it but now my body is like girllllllllll.
0:15

i went to a party by myself last night. got on the stage, danced, threw my leg up and hit a split. the crowd enjoyed it but now my body is like girllllllllll.

holistic mami✨

142,453 просмотров • 7 месяцев назад

We just launched Claude Sonnet 4.5 and “Imagine with Claude”, a 5-day preview that creates dynamic experiences for you in real-time (and the first product I’ve worked on at Anthropic!). It pioneers the concept of “model-as-backend”, using a model to not only generate interfaces on the fly but also power all the functionality behind it – all made possible by our newest model, Claude Sonnet 4.5. Here’s a "choose your own adventure" version of my founder journey, as retold by Imagine with Claude:
0:17

We just launched Claude Sonnet 4.5 and “Imagine with Claude”, a 5-day preview that creates dynamic experiences for you in real-time (and the first product I’ve worked on at Anthropic!). It pioneers the concept of “model-as-backend”, using a model to not only generate interfaces on the fly but also power all the functionality behind it – all made possible by our newest model, Claude Sonnet 4.5. Here’s a "choose your own adventure" version of my founder journey, as retold by Imagine with Claude:

Sean Strong

115,693 просмотров • 10 месяцев назад

should i keep working on this? I was just doing it for fun but I'm getting kinda bored and stressed from it lol
0:12

should i keep working on this? I was just doing it for fun but I'm getting kinda bored and stressed from it lol

JOY of 3D

20,855 просмотров • 10 месяцев назад

Happy Black History Month! It may be the last day but it’s never over! I’m proud to be a black woman making music and subverting expectations. The world is hard to live in right now but I’ll follow my dreams and keep moving forward, always loving myself and where I came from!
0:19

Happy Black History Month! It may be the last day but it’s never over! I’m proud to be a black woman making music and subverting expectations. The world is hard to live in right now but I’ll follow my dreams and keep moving forward, always loving myself and where I came from!

PAiDA👽new song out now!

26,966 просмотров • 1 год назад

Test-time learning has a catch that scaling alone does not solve: the state a system writes to in order to learn is also the record a user must inspect to audit it. Optimize it only for learning, and auditability usually degrades. Blog: Part I treats the learning state and the audit record as the same object. This starts the ADAPT series, exploring the architecture section by section.
0:11

Test-time learning has a catch that scaling alone does not solve: the state a system writes to in order to learn is also the record a user must inspect to audit it. Optimize it only for learning, and auditability usually degrades. Blog: Part I treats the learning state and the audit record as the same object. This starts the ADAPT series, exploring the architecture section by section.

Rei

32,519 просмотров • 8 дней назад

Talking about bicycle scene’s filming 🚲 Junho: “It was comfortable, right?” YoonA: “It was comfortable, but since I also had to hold my skirt, I put strength on my legs and it felt like I was exercising.” Junho: “I know this because I’ve ridden bicycles a lot, but that must be difficult.” YoonA: “(The ride) was stable.” Junho: “I just finished doing cardio.”
0:18

Talking about bicycle scene’s filming 🚲 Junho: “It was comfortable, right?” YoonA: “It was comfortable, but since I also had to hold my skirt, I put strength on my legs and it felt like I was exercising.” Junho: “I know this because I’ve ridden bicycles a lot, but that must be difficult.” YoonA: “(The ride) was stable.” Junho: “I just finished doing cardio.”

😵‍💫🍬

40,994 просмотров • 3 лет назад

Pro-Palestine supporter proudly posted a video stealing coffee from Starbucks, as if that somehow supports people in Gaza. This demonstrates that the left is not only willing to engage in criminal behavior, but also quick to justify it by invoking their cause.
0:15

Pro-Palestine supporter proudly posted a video stealing coffee from Starbucks, as if that somehow supports people in Gaza. This demonstrates that the left is not only willing to engage in criminal behavior, but also quick to justify it by invoking their cause.

I Meme Therefore I Am 🇺🇸

39,851 просмотров • 9 месяцев назад

Ummmm no WALLHACK© sp005 review anymore? I didn’t slam it, drop it, or anything. I noticed the small break on the edge today, haven’t moved this pad for over a week. My only thought is it’s my keyboard running into it. And I pushed my keyboard too hard into the pad and it broke. I tried to pick it up right after seeing that and it exploded in my face making this much larger crack. The pad basically isn’t useable anymore.
0:13

Ummmm no WALLHACK© sp005 review anymore? I didn’t slam it, drop it, or anything. I noticed the small break on the edge today, haven’t moved this pad for over a week. My only thought is it’s my keyboard running into it. And I pushed my keyboard too hard into the pad and it broke. I tried to pick it up right after seeing that and it exploded in my face making this much larger crack. The pad basically isn’t useable anymore.

Mighty+

41,093 просмотров • 3 месяцев назад

💌 260202 1:11 AM 🎥 🦊 i ate jjimdak with hyeongjun 🦊 and i went for a walk by myself 🦊 it was snowing so i filmed it 🦊 and i came home and sent it 🦊 have a pitch black dream 💜
0:12

💌 260202 1:11 AM 🎥 🦊 i ate jjimdak with hyeongjun 🦊 and i went for a walk by myself 🦊 it was snowing so i filmed it 🦊 and i came home and sent it 🦊 have a pitch black dream 💜

🦊

25,003 просмотров • 6 месяцев назад

With it being St. Patrick’s Day this week I thought it was only right to put you on to some Irish talent KhakiKid is taking Ireland by storm right now, and I can totally see why
0:16

With it being St. Patrick’s Day this week I thought it was only right to put you on to some Irish talent KhakiKid is taking Ireland by storm right now, and I can totally see why

UK Rap Daily

58,151 просмотров • 1 год назад

🐰 : “Going to watch a movie by myself is so much fun, a bit thrilling, and not bad at all, I think. I should go by myself every now and then; it was really enjoyable” So glad he's enjoying his free time 🥹
0:17

🐰 : “Going to watch a movie by myself is so much fun, a bit thrilling, and not bad at all, I think. I should go by myself every now and then; it was really enjoyable” So glad he's enjoying his free time 🥹

Daily JK Praise 

55,841 просмотров • 13 дней назад