Loading video...

Video Failed to Load

There was a problem loading this video. This could be due to a temporary network issue or the video might be unavailable.

Presenting Unsupervised Actuator Nets (UANs) that push the limits of agile whole-body control without the need for reward shaping! ⚡️ UANs reduce the sim2real gap in robot's motors removing the need for reward design to bridge the sim2real gap. ⚡️ A pre-trained whole-body controller uses reference motion as a... show more

Pulkit Agrawal

16,675 subscribers

11,196 views • 1 year ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

3 Comments

Pulkit Agrawal1 year ago

Pushing the limits requires brave souls!

NICE1 year ago

Stay competitive by balancing cutting-edge AI with automation tools. Forrester shows how.

Ray Kuo1 year ago

cool!

Related Videos

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.

TWIST: a real-time teleoperation system for humanoid robots to mimic whole-body motions. Reference motion data is generated by retargeting human motion-capture data to the robot. Then, the controller is trained in simulation using reinforcement learning and behavior cloning.

The Humanoid Hub

46,715 views • 1 year ago

This work is not about a new technique. GMT (General Motion Tracking) shows good engineering practices that you can actually train a single unified whole-body control policy for all agile motion, and it works in the real world, directly with sim2real without adaptation. This is different from many existing works that trains/adapts one individual policy for each specific motion. Instead of just showing the beautiful motions/dances, this is actually practical. Because now you can convert all types of commands into real actions in real time without tuning anything. Visit to check out how we handle the following challenges: ✅ Partial Observability. ✅ Hardware Limitations. ✅ Unbalanced Data Distribution. ✅ Model Expressiveness. Work led by Zixuan Chen Mazeyu Ji

This work is not about a new technique. GMT (General Motion Tracking) shows good engineering practices that you can actually train a single unified whole-body control policy for all agile motion, and it works in the real world, directly with sim2real without adaptation. This is different from many existing works that trains/adapts one individual policy for each specific motion. Instead of just showing the beautiful motions/dances, this is actually practical. Because now you can convert all types of commands into real actions in real time without tuning anything. Visit to check out how we handle the following challenges: ✅ Partial Observability. ✅ Hardware Limitations. ✅ Unbalanced Data Distribution. ✅ Model Expressiveness. Work led by Zixuan Chen Mazeyu Ji

Xiaolong Wang

83,261 views • 1 year ago

Ready to play with your whole body in solo? ✨ Full Body Tracking is BACK in Just Dance 2026 Edition, now on all consoles! 🕺💃 All you need is the free Just Dance Controller App! With Camera Controller Mode, your body is the controller. No limits. No excuses. Just Dance.

Ready to play with your whole body in solo? ✨ Full Body Tracking is BACK in Just Dance 2026 Edition, now on all consoles! 🕺💃 All you need is the free Just Dance Controller App! With Camera Controller Mode, your body is the controller. No limits. No excuses. Just Dance.

Just Dance 2026 Edition

63,875 views • 9 months ago

Motion design takes things to a whole different level Great work by JΛVIΞR for Halaska Studio ⚡️

Motion design takes things to a whole different level Great work by JΛVIΞR for Halaska Studio ⚡️

Chris Halaska

28,496 views • 1 year ago

Like LLMs, autoregressive transformers for action tokens need a reasoning layer to reduce hallucinations and boost reliability. Grounding this layer in the physics of the action space using DVBFs makes for scalable, task-agnostic training—far simpler than creating RL reward functions for each task. Learn more about our novel approach in the video below 👇🏼

Like LLMs, autoregressive transformers for action tokens need a reasoning layer to reduce hallucinations and boost reliability. Grounding this layer in the physics of the action space using DVBFs makes for scalable, task-agnostic training—far simpler than creating RL reward functions for each task. Learn more about our novel approach in the video below 👇🏼

Sankaet

52,565 views • 1 year ago

🚀 Introducing SoftMimic: Compliant humanoids for an interactive future — bringing humanoids into the real world 🤝 🔗 Current humanoids collapse when they touch the world — they can’t handle contact or deviation from their reference motion 😭 😎SoftMimic learns to deviate gracefully from motion references. It natively embraces contact and forces — enabling safer, more robust robots that generalize better and interact gently. Work led by Gabe Margolis Michelle and Nolan Fey

🚀 Introducing SoftMimic: Compliant humanoids for an interactive future — bringing humanoids into the real world 🤝 🔗 Current humanoids collapse when they touch the world — they can’t handle contact or deviation from their reference motion 😭 😎SoftMimic learns to deviate gracefully from motion references. It natively embraces contact and forces — enabling safer, more robust robots that generalize better and interact gently. Work led by Gabe Margolis Michelle and Nolan Fey

Pulkit Agrawal

40,167 views • 8 months ago

Mobile-TeleVision follows our previous idea on upper-body and lower-body separation for control: 1⃣We train RL for lower-body locomotion and standing; 2⃣We use IK and retargeting directly for upper-body for high-precision and smooth manipulation. We avoid using RL for upper-body motion tracking. This is because fundamentally it is hard to avoid errors with RL tracking, and sim2real gap brings another gap. At this point, we still find using IK is more practical and precise.

Mobile-TeleVision follows our previous idea on upper-body and lower-body separation for control: 1⃣We train RL for lower-body locomotion and standing; 2⃣We use IK and retargeting directly for upper-body for high-precision and smooth manipulation. We avoid using RL for upper-body motion tracking. This is because fundamentally it is hard to avoid errors with RL tracking, and sim2real gap brings another gap. At this point, we still find using IK is more practical and precise.

Xiaolong Wang

32,662 views • 1 year ago

“It’s a global problem with a local solution.” Pierre Poilievre says that by removing the carbon tax and replacing it with a “tax reward” for companies to reduce their emissions, producing things in Canada will reduce pollution globally.

“It’s a global problem with a local solution.” Pierre Poilievre says that by removing the carbon tax and replacing it with a “tax reward” for companies to reduce their emissions, producing things in Canada will reduce pollution globally.

Juno News

36,337 views • 1 year ago

Ready to experience Temerario in a whole new dimension? Exclusive digital liveries are now live in The Crew. Take the wheel and push the limits of performance in the virtual world. Discover more:

Ready to experience Temerario in a whole new dimension? Exclusive digital liveries are now live in The Crew. Take the wheel and push the limits of performance in the virtual world. Discover more:

Lamborghini

23,953 views • 6 months ago

Leon Thomas says meditation has changed his life: "In your mental health journey, to bridge the gap meditation trained my self conscious to envision myself you need that quiet time in the morning and at night for affirmation for your goals"

Leon Thomas says meditation has changed his life: "In your mental health journey, to bridge the gap meditation trained my self conscious to envision myself you need that quiet time in the morning and at night for affirmation for your goals"

Ahmed/The Ears/IG: BigBizTheGod 🇸🇴

134,303 views • 9 months ago

The only thing holding you back is your inability to do the work that you know you need to do, without witness or immediate reward for a long enough period of time.

The only thing holding you back is your inability to do the work that you know you need to do, without witness or immediate reward for a long enough period of time.

New Mentalities

122,952 views • 1 year ago

There’s a massive gap between my perception of Nate Ament and the teams that are valuing him as a T10 pick in this draft. The Nets & Bucks are setting themselves up to make the same exact mistakes that have led to their underwhelming draft performance in the last 5 years

There’s a massive gap between my perception of Nate Ament and the teams that are valuing him as a T10 pick in this draft. The Nets & Bucks are setting themselves up to make the same exact mistakes that have led to their underwhelming draft performance in the last 5 years

David Lee

87,907 views • 19 days ago

Your heart is the center of wholeness – the bridge to the divine within. With loving attention and intention, you can connect its healing power to a part of your body that needs to become more whole. Experience “Love Your Body”

Your heart is the center of wholeness – the bridge to the divine within. With loving attention and intention, you can connect its healing power to a part of your body that needs to become more whole. Experience “Love Your Body”

Dr Joe Dispenza

16,697 views • 1 year ago

This gotta be one of the best takes on the whole "flying over the stage" argument. That type of stuff absolutely has a place in Sonic games as a reward for high skill play and it does not in any way invalidate the level design

This gotta be one of the best takes on the whole "flying over the stage" argument. That type of stuff absolutely has a place in Sonic games as a reward for high skill play and it does not in any way invalidate the level design

Proto arts

156,728 views • 11 months ago

Will the wheel turn in your favor?! 🤩 For a chance to win BIG… ⚡️ Snap a screenshot of the reward you want ⚡️ Drop it down below (along with your Game ID) ⚡️ If you’re lucky, YOU could walk away with an amazing prize! 🏆 ⚡️ More screenshots = more chances to win! #TravelTown

Will the wheel turn in your favor?! 🤩 For a chance to win BIG… ⚡️ Snap a screenshot of the reward you want ⚡️ Drop it down below (along with your Game ID) ⚡️ If you’re lucky, YOU could walk away with an amazing prize! 🏆 ⚡️ More screenshots = more chances to win! #TravelTown

Travel Town Game

52,491 views • 1 year ago

We are excited to share our #CORL2024 paper on learning quadrotor obstacle avoidance from the visual stream of a single #eventcamera! Trained entirely in simulation! We demonstrate obstacle avoidance both in the dark and in a forest up to 5m/s. PDF: Video: Project page: Event cameras are sensors that output per-pixel-level intensity changes at microsecond latency resolution; they feature nearly zero motion blur and high dynamic range but produce a very large volume of events under significant ego-motion and further lack a high-fidelity continuous-time sensor model in simulation, making direct #sim2real transfer not possible. By leveraging depth prediction as a pretext task, we pre-train a reactive obstacle avoidance policy with “approximated” simulated events and then fine-tune the perception component with limited events-and-depth real-world data. This technique bridges the sim2real gap for #eventcameras! As at the current state, there is no continuous-time sensor model for event cameras, we hope that this work can finally spur future research leveraging simulation for training event-vision-based policies to create faster, agile robots! Kudos to Anish Bhattacharya, @marcocannic, Vijay Kumar Nikolai Matni UZH Science University of Zurich UZH Space Hub UZH IfI European Research Council (ERC) GRASP Laboratory Penn Engineering

We are excited to share our #CORL2024 paper on learning quadrotor obstacle avoidance from the visual stream of a single #eventcamera! Trained entirely in simulation! We demonstrate obstacle avoidance both in the dark and in a forest up to 5m/s. PDF: Video: Project page: Event cameras are sensors that output per-pixel-level intensity changes at microsecond latency resolution; they feature nearly zero motion blur and high dynamic range but produce a very large volume of events under significant ego-motion and further lack a high-fidelity continuous-time sensor model in simulation, making direct #sim2real transfer not possible. By leveraging depth prediction as a pretext task, we pre-train a reactive obstacle avoidance policy with “approximated” simulated events and then fine-tune the perception component with limited events-and-depth real-world data. This technique bridges the sim2real gap for #eventcameras! As at the current state, there is no continuous-time sensor model for event cameras, we hope that this work can finally spur future research leveraging simulation for training event-vision-based policies to create faster, agile robots! Kudos to Anish Bhattacharya, @marcocannic, Vijay Kumar Nikolai Matni UZH Science University of Zurich UZH Space Hub UZH IfI European Research Council (ERC) GRASP Laboratory Penn Engineering

Davide Scaramuzza

17,219 views • 1 year ago

"The work is the reward - not being recognized for the work." Players and teams that work hard are rewarded with opportunities to work even harder. Relentless never rests. Champions don’t need recognition. They need the next challenge.

"The work is the reward - not being recognized for the work." Players and teams that work hard are rewarded with opportunities to work even harder. Relentless never rests. Champions don’t need recognition. They need the next challenge.

The Winning Difference

15,684 views • 7 months ago

When you have done a lot of exercise and now you need to restore the sugars back in your body. You have to eat the donuts the whole box

When you have done a lot of exercise and now you need to restore the sugars back in your body. You have to eat the donuts the whole box

iamtakudzwa

14,799 views • 1 month ago