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.

A new Science Robotics study highlights a reinforcement learning-based, zero-shot approach that uses motion capture and animation to train humanoid and four-legged robots to perform acrobatics.

Science Magazine

4,774,724 subscribers

56,242 views • 8 days ago •via X (Twitter)

Science & Technology

Anya Rossi• Live Now

Private livecam show

0 Comments

No comments available

Comments from the original post will appear here

Related Videos

A new Science Robotics study highlights a reinforcement learning–based framework that integrates perceptual uncertainty into policy learning to equip humanoid robots with reactive soccer skills and dynamic behaviors.

A new Science Robotics study highlights a reinforcement learning–based framework that integrates perceptual uncertainty into policy learning to equip humanoid robots with reactive soccer skills and dynamic behaviors.

Science Magazine

56,716 views • 25 days ago

A new approach uses diffused orientation fields to help robots transfer skills across curved objects of diverse shapes and sizes, enabling them to peel and slice varied fruits and vegetables. Learn more in Science Robotics:

A new approach uses diffused orientation fields to help robots transfer skills across curved objects of diverse shapes and sizes, enabling them to peel and slice varied fruits and vegetables. Learn more in Science Robotics:

Science Magazine

18,401 views • 4 months ago

A new Science Robotics study highlights exoskeletons that allow physical therapists and patients recovering from stroke to “feel” each other’s movements as they walk on separate treadmills, enabling nuanced gait correction.

A new Science Robotics study highlights exoskeletons that allow physical therapists and patients recovering from stroke to “feel” each other’s movements as they walk on separate treadmills, enabling nuanced gait correction.

Science Magazine

40,401 views • 2 months ago

Just released five robotics simulation environments we have been using to train reinforcement learning from vision and touch! Very easy to install. Powered by Google DeepMind MuJoCo, Gymnasium, and robosuite. GitHub link:

Just released five robotics simulation environments we have been using to train reinforcement learning from vision and touch! Very easy to install. Powered by Google DeepMind MuJoCo, Gymnasium, and robosuite. GitHub link:

Carlo Sferrazza

24,389 views • 2 years ago

Ever had a perfect image in mind that a text-to-image model just couldn't capture? Our new reinforcement learning agent, PASTA, turns image generation into a collaborative conversation, learning your style to bring your vision to life. Learn how it works:

Ever had a perfect image in mind that a text-to-image model just couldn't capture? Our new reinforcement learning agent, PASTA, turns image generation into a collaborative conversation, learning your style to bring your vision to life. Learn how it works:

Google Research

25,860 views • 11 months ago

Generalist robots need multimodal learning—fusing vision, touch, and body motion to handle unstructured worlds. 🦾 NVIDIA Isaac Lab is a GPU-accelerated, open-source simulation framework for robot learning that unifies physics, rendering, sensing, and parallel computing to train robot policies at scale. Learn more 🔗 #NVIDIAResearch

Generalist robots need multimodal learning—fusing vision, touch, and body motion to handle unstructured worlds. 🦾 NVIDIA Isaac Lab is a GPU-accelerated, open-source simulation framework for robot learning that unifies physics, rendering, sensing, and parallel computing to train robot policies at scale. Learn more 🔗 #NVIDIAResearch

NVIDIA Robotics

18,568 views • 7 months ago

New open-source library speeds up whole-body imitation learning. Train the policy below in under 20 minutes ⬇️ LocoMuJoCo2 brings biomechanics and humanoid robotics into one clean, high-speed benchmark. Built for researchers who want to train smarter and faster. What it offers ✅ 22K+ motion trajectories per humanoid model ✅ Supports both MuJoCo and MJX ✅ JAX-based PPO, GAIL, AMP, DeepMimic implementations ✅ Plug-and-play robot2robot retargeting pipeline Whether you’re building humanoids or studying human motion, this toolkit saves you weeks of setup. Github: Thank you, Firas Al-Hafez for sharing this great work!

New open-source library speeds up whole-body imitation learning. Train the policy below in under 20 minutes ⬇️ LocoMuJoCo2 brings biomechanics and humanoid robotics into one clean, high-speed benchmark. Built for researchers who want to train smarter and faster. What it offers ✅ 22K+ motion trajectories per humanoid model ✅ Supports both MuJoCo and MJX ✅ JAX-based PPO, GAIL, AMP, DeepMimic implementations ✅ Plug-and-play robot2robot retargeting pipeline Whether you’re building humanoids or studying human motion, this toolkit saves you weeks of setup. Github: Thank you, Firas Al-Hafez for sharing this great work!

Ilir Aliu

23,197 views • 1 year ago

With Dual Override, this approach appears to have been revisited. The new model features longer proportions to support more expressive movement and animation, an approach Capcom previously explored with the motion-capture model shown earlier this year at the Great Capcom Exhibition.

With Dual Override, this approach appears to have been revisited. The new model features longer proportions to support more expressive movement and animation, an approach Capcom previously explored with the motion-capture model shown earlier this year at the Great Capcom Exhibition.

Brian

14,904 views • 9 months ago

Bringing new meaning to the word NeRD at #CoRL2025. 🤓 Neural Robot Dynamics (NeRD) bridges the sim-to-real gap with a hybrid of neural modeling and analytical simulation, unlocking zero-shot transfer and continuous learning loops for optimal robotics training. Read more & try it in Newton 🤖

Bringing new meaning to the word NeRD at #CoRL2025. 🤓 Neural Robot Dynamics (NeRD) bridges the sim-to-real gap with a hybrid of neural modeling and analytical simulation, unlocking zero-shot transfer and continuous learning loops for optimal robotics training. Read more & try it in Newton 🤖

NVIDIA Robotics

17,556 views • 11 months ago

Working on a new topic and hardware for a while: Feet Language Action models. VLAs teach robots to act through vision and language FLAs teach legged robots to act through contact Given a task, terrain, body state, and instruction - decide where, when, and how to make contact

Working on a new topic and hardware for a while: Feet Language Action models. VLAs teach robots to act through vision and language FLAs teach legged robots to act through contact Given a task, terrain, body state, and instruction - decide where, when, and how to make contact

Irvin (in Japan 🇯🇵)

14,004 views • 2 months ago

System ID for legged robots is hard: (1) Discontinuous dynamics and (2) many parameters to identify and hard to "excite" them. SPI-Active is a general tool for legged robot system ID. Key ideas: (1) massively parallel sampling-based optimization, (2) structured parameter space, and (3) active exploration based on Fisher Information to collect the most informative data in real. SPI-Active provides an accurate robot model and effectively reduces the sim2real gap. In sim2real policy learning setting, it outperforms baselines by 42-63% in various quadruped & humanoid tasks. Led by Nikhil Sobanbabu Guanqi He

System ID for legged robots is hard: (1) Discontinuous dynamics and (2) many parameters to identify and hard to "excite" them. SPI-Active is a general tool for legged robot system ID. Key ideas: (1) massively parallel sampling-based optimization, (2) structured parameter space, and (3) active exploration based on Fisher Information to collect the most informative data in real. SPI-Active provides an accurate robot model and effectively reduces the sim2real gap. In sim2real policy learning setting, it outperforms baselines by 42-63% in various quadruped & humanoid tasks. Led by Nikhil Sobanbabu Guanqi He

Guanya Shi

21,442 views • 1 year ago

26 years of humanoid robotics in 16 seconds. In 2000 keeping a biped robot upright was already an achievement. In 2026 humanoids are exchanging kicks, maintaining balance and recovering inside a robot combat arena. The progress in actuators, control systems and mechanical stability is impossible to ignore. What will humanoid robots be doing in 2036? #HumanoidRobots #Robotics #PhysicalAI

26 years of humanoid robotics in 16 seconds. In 2000 keeping a biped robot upright was already an achievement. In 2026 humanoids are exchanging kicks, maintaining balance and recovering inside a robot combat arena. The progress in actuators, control systems and mechanical stability is impossible to ignore. What will humanoid robots be doing in 2036? #HumanoidRobots #Robotics #PhysicalAI

Techniahqrobot | humanoid robots

41,728 views • 1 month ago

Using attention-based map encoding, we enable robots to traverse extremely sparse terrains. We are now developing a new architecture that explicitly plans contact points, rather than latent representations, enabling whole-body contact planning for humanoid robots. Stay tuned.

Using attention-based map encoding, we enable robots to traverse extremely sparse terrains. We are now developing a new architecture that explicitly plans contact points, rather than latent representations, enabling whole-body contact planning for humanoid robots. Stay tuned.

Junzhe (JJ) He

23,555 views • 8 months ago