Video wird geladen...

Video konnte nicht geladen werden

Zur Startseite

population based training (PBT) is underrated for pushing scale and getting better results in GPU-accelerated RL. Our new work DexPBT lead by Aleksei Petrenko shows how it can be used to train highly dexterous hand-arm manipulation in up to 46 DoF systems.

arthurallshire's profile picture

Arthur Allshire

2,553 subscribers

13,569 Aufrufe • vor 3 Jahren •via X (Twitter)

LIVE

Anya Rossi• Live Now

Private livecam show

8 Kommentare

Profilbild von Arthur Allshire
Arthur Allshirevor 3 Jahren

PBT runs a genetic algorithm on top of basic RL. This does multiple things: gives more diversity in the initial policies (reducing intra-seed variability), helps find better hyperparameters, and counteracts the instability that plagues harder RL environments even with good HPs

Profilbild von Arthur Allshire
Arthur Allshirevor 3 Jahren

one nice thing about DexPBT's implementation of PBT is it is decentralised -- only need one shared folder between runs. Slurm, local multi-GPU machine, NGC, other cluster software -- our implementation doesn't care.

Profilbild von Arthur Allshire
Arthur Allshirevor 3 Jahren

this work gives me hope for RL in increasingly complex embodiments and scenarios. PBT works in other robotic environments... stay tuned ;) video - paper - thanks to collaborators @petrenko_ai @ankurhandos @viktor_m81 @gavrielstate

Profilbild von Maggie
Maggievor 3 Jahren

@petrenko_ai damn this is so cool

Profilbild von Jonathan A. Michaels
Jonathan A. Michaelsvor 3 Jahren

@petrenko_ai Very cool

Profilbild von David Hoeller
David Hoellervor 3 Jahren

@petrenko_ai That’s probably the most impressive manipulation work I have ever seen, congrats!

Profilbild von Dennis Anthony
Dennis Anthonyvor 3 Jahren

@petrenko_ai Super cool work Arthur!!

Profilbild von Zhihao Liu
Zhihao Liuvor 3 Jahren

@petrenko_ai 👍

Ähnliche Videos

How to learn dexterous manipulation for any robot hand from a single human demonstration? Check out DexMachina, our new RL algorithm that learns long-horizon, bimanual dexterous policies for a variety of dexterous hands, articulated objects, and complex motions.
0:43

How to learn dexterous manipulation for any robot hand from a single human demonstration? Check out DexMachina, our new RL algorithm that learns long-horizon, bimanual dexterous policies for a variety of dexterous hands, articulated objects, and complex motions.

Mandi Zhao

122,281 Aufrufe • vor 1 Jahr

Sim2Real RL for Vision-Based Dexterous Manipulation on Humanoids TLDR - we train a humanoid robot with two multifingered hands to perform a range of dexterous manipulation tasks robust generalization and high performance without human demonstration :D
0:20

Sim2Real RL for Vision-Based Dexterous Manipulation on Humanoids TLDR - we train a humanoid robot with two multifingered hands to perform a range of dexterous manipulation tasks robust generalization and high performance without human demonstration :D

Toru

49,892 Aufrufe • vor 1 Jahr

🔥 #ICRA2026 Best Paper Finalist The era of "robot VLA = single-arm gripper" is ending. Introducing Dexora — the first open-source Vision-Language-Action system for dual-arm, dual-hand, 36-DoF dexterous manipulation. 🦾 Dual Arms 🖐️ Dual Hands 🎯 36 DoF Control 🌍 Open Source Trained on: • 100K simulated trajectories • 10K real-world demonstrations Dexora achieves: ✓ 90%+ success on basic manipulation ✓ Strong dexterous manipulation performance ✓ Cross-embodiment generalization Our key hypothesis: Train on the hardest embodiment. Transfer to simpler robots later. Instead of scaling up gripper policies, we train directly in the most expressive action space and project downward to simpler embodiments. This may be a practical path toward universal robot controllers. 🎥 Demos: 📄 Paper:
2:58

🔥 #ICRA2026 Best Paper Finalist The era of "robot VLA = single-arm gripper" is ending. Introducing Dexora — the first open-source Vision-Language-Action system for dual-arm, dual-hand, 36-DoF dexterous manipulation. 🦾 Dual Arms 🖐️ Dual Hands 🎯 36 DoF Control 🌍 Open Source Trained on: • 100K simulated trajectories • 10K real-world demonstrations Dexora achieves: ✓ 90%+ success on basic manipulation ✓ Strong dexterous manipulation performance ✓ Cross-embodiment generalization Our key hypothesis: Train on the hardest embodiment. Transfer to simpler robots later. Instead of scaling up gripper policies, we train directly in the most expressive action space and project downward to simpler embodiments. This may be a practical path toward universal robot controllers. 🎥 Demos: 📄 Paper:

Hao Zhao

17,583 Aufrufe • vor 5 Monaten

How to chain multiple dexterous skills to tackle complex long-horizon manipulation tasks? Imagine retrieving a LEGO block from a pile, rotating it in-hand, and inserting it at the desired location to build a structure. Introducing our new work - Sequential Dexterity 🧵👇
0:52

How to chain multiple dexterous skills to tackle complex long-horizon manipulation tasks? Imagine retrieving a LEGO block from a pile, rotating it in-hand, and inserting it at the desired location to build a structure. Introducing our new work - Sequential Dexterity 🧵👇

Chen Wang

159,920 Aufrufe • vor 3 Jahren

Want to scale RL with your shiny new GPU? 🚀 In our ICML24 Oral we find that RL algorithms hit a barrier when data is scaled up. Our new algorithm, SAPG, proposes a simple fix. It scales to 25k envs and solves hard tasks where PPO makes no progress. 1/n
1:24

Want to scale RL with your shiny new GPU? 🚀 In our ICML24 Oral we find that RL algorithms hit a barrier when data is scaled up. Our new algorithm, SAPG, proposes a simple fix. It scales to 25k envs and solves hard tasks where PPO makes no progress. 1/n

Ananye Agarwal

64,610 Aufrufe • vor 2 Jahren

We talked to Ritvik Singh about how you can train sim-to-real dexterous manipulation policies using NVIDIA Isaac. This robot is grasping objects using pure RGB stereo: take in images from a camera pair and predict what to do, all without training in the real world.
1:08

We talked to Ritvik Singh about how you can train sim-to-real dexterous manipulation policies using NVIDIA Isaac. This robot is grasping objects using pure RGB stereo: take in images from a camera pair and predict what to do, all without training in the real world.

Chris Paxton

20,067 Aufrufe • vor 1 Jahr

📢📢📢 Excited to release ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR25). 🤏🤙✌️With ManipTrans, we can transfer dexterous manipulation skills into robotic hands in simulation and deploy them on a real robot, using a residual policy learned for dex manipulation. 🤖🤖🤖The video below illustrates how the MoCap data can be transferred to Inspire, Shadow, Xhand, Allegro, and Mano. With ManipTrans, we can scale up dex manip data greatly with minimal effort. For more details, please check our -webpage: -code: -huggingface:
1:59

📢📢📢 Excited to release ManipTrans: Efficient Dexterous Bimanual Manipulation Transfer via Residual Learning (CVPR25). 🤏🤙✌️With ManipTrans, we can transfer dexterous manipulation skills into robotic hands in simulation and deploy them on a real robot, using a residual policy learned for dex manipulation. 🤖🤖🤖The video below illustrates how the MoCap data can be transferred to Inspire, Shadow, Xhand, Allegro, and Mano. With ManipTrans, we can scale up dex manip data greatly with minimal effort. For more details, please check our -webpage: -code: -huggingface:

Siyuan Huang

20,983 Aufrufe • vor 1 Jahr

Below is our open-source arm assembled by one of the discord members! Also, today Physical Intelligence shared their incredible top-notch robotics model Pi0. And I just want to remind you that you can use our arm together with this model and with LeRobot The arm has the same DOF as the arm used by Pi0, but has smaller size and cost starts from 200$ per arm. You can use better motors like we did in our demos, then it will be more precise and stable but the price will be about $600 per arm. Printable files, purchase list, and help with the assembly - in our discord channel.
0:25

Below is our open-source arm assembled by one of the discord members! Also, today Physical Intelligence shared their incredible top-notch robotics model Pi0. And I just want to remind you that you can use our arm together with this model and with LeRobot The arm has the same DOF as the arm used by Pi0, but has smaller size and cost starts from 200$ per arm. You can use better motors like we did in our demos, then it will be more precise and stable but the price will be about $600 per arm. Printable files, purchase list, and help with the assembly - in our discord channel.

Igor Kulakov (MicroFactory)

53,938 Aufrufe • vor 1 Jahr

Robostral can now follow natural language instructions. It responds to voice commands and pointing. It is also getting better at fine-grained manipulation where precise force control matters. It generalizes to new objects and tasks not present in the training data.
1:07

Robostral can now follow natural language instructions. It responds to voice commands and pointing. It is also getting better at fine-grained manipulation where precise force control matters. It generalizes to new objects and tasks not present in the training data.

Olivier Duchenne

52,067 Aufrufe • vor 5 Monaten

This is a demonstration video of in-hand object rotation with new Torobo Hand. By leveraging large-scale parallel reinforcement learning in NVIDIA Isaac Sim, the learned policy can perform object manipulation in MuJoCo and on the real robot without any additional training.
1:09

This is a demonstration video of in-hand object rotation with new Torobo Hand. By leveraging large-scale parallel reinforcement learning in NVIDIA Isaac Sim, the learned policy can perform object manipulation in MuJoCo and on the real robot without any additional training.

東京ロボティクス株式会社

10,951 Aufrufe • vor 1 Jahr

From day one, mimic has been focused on a single goal: general-purpose dexterous manipulation. Today we're proud to announce the mimic hand M1 and the mimic wearable U1. We believe the only way to solve dexterous manipulation at scale is by going full-stack at the frontier of physical AI, building every layer ourselves around one fixed point, the human hand. The M1 is a highly backdrivable, tendon-driven hand that covers the full range of human capability, from heavy payloads to fine manipulation.
1:43

From day one, mimic has been focused on a single goal: general-purpose dexterous manipulation. Today we're proud to announce the mimic hand M1 and the mimic wearable U1. We believe the only way to solve dexterous manipulation at scale is by going full-stack at the frontier of physical AI, building every layer ourselves around one fixed point, the human hand. The M1 is a highly backdrivable, tendon-driven hand that covers the full range of human capability, from heavy payloads to fine manipulation.

mimic

94,354 Aufrufe • vor 3 Monaten

Excited to share our new blog: Scaling Video Training with Parallelism We discuss how parallelism helps scale long-video training systems, for both understanding and generation.
1:00

Excited to share our new blog: Scaling Video Training with Parallelism We discuss how parallelism helps scale long-video training systems, for both understanding and generation.

Yukang Chen

12,802 Aufrufe • vor 5 Monaten

Just last night, in response to Secretary of War Pete Hegseth’s establishment of Autonomous Warfare Command, the U.S. Army announced Futures and Autonomous Systems Command (FASCOM). This is how we couple all of the great work that's happening with our vast resources and centralize them so we can scale and coordinate. We're pushing autonomy down into our units, but when it comes to acquisition and making sure our left hand knows what the right hand is doing, this happens in FASCOM, and feeds up into AutoWarCom and across the services. Live at #AUSA2026
0:40

Just last night, in response to Secretary of War Pete Hegseth’s establishment of Autonomous Warfare Command, the U.S. Army announced Futures and Autonomous Systems Command (FASCOM). This is how we couple all of the great work that's happening with our vast resources and centralize them so we can scale and coordinate. We're pushing autonomy down into our units, but when it comes to acquisition and making sure our left hand knows what the right hand is doing, this happens in FASCOM, and feeds up into AutoWarCom and across the services. Live at #AUSA2026

Acting Secretary of the Army

11,479 Aufrufe • vor 23 Tagen

This is going to be the year of hands and dexterous manipulation :) A sneak-peak into what's been cooking in our lab:
1:08

This is going to be the year of hands and dexterous manipulation :) A sneak-peak into what's been cooking in our lab:

Pulkit Agrawal

22,883 Aufrufe • vor 2 Jahren

Just collecting manipulation data isn’t enough for robots - they need to be able to move around in the world, which has a whole different set of challenges from pure manipulation. And bringing navigation and manipulation together in a single framework is even more challenging. Enter HERMES, from Zhecheng Yuan and Tianming Wei. This is a four-stage process in which human videos are used to set up an RL sim-to-real training pipeline in order to overcome differences between robot and human kinematics, and used together with a navigation foundation model to move around in a variety of environments. To learn more, join us as Zhecheng Yuan and Tianming Wei tell us about how they built their system to perform mobile dexterous manipulation from human videos in a variety of environments. Watch Episode #45 of RoboPapers today, hosted by Michael Cho - Rbt/Acc and Chris Paxton!
1:01:08

Just collecting manipulation data isn’t enough for robots - they need to be able to move around in the world, which has a whole different set of challenges from pure manipulation. And bringing navigation and manipulation together in a single framework is even more challenging. Enter HERMES, from Zhecheng Yuan and Tianming Wei. This is a four-stage process in which human videos are used to set up an RL sim-to-real training pipeline in order to overcome differences between robot and human kinematics, and used together with a navigation foundation model to move around in a variety of environments. To learn more, join us as Zhecheng Yuan and Tianming Wei tell us about how they built their system to perform mobile dexterous manipulation from human videos in a variety of environments. Watch Episode #45 of RoboPapers today, hosted by Michael Cho - Rbt/Acc and Chris Paxton!

RoboPapers

15,092 Aufrufe • vor 11 Monaten

Advancing dexterous manipulation through scalable visual sim-to-real transfer. We are excited to share our RSS paper, “ViserDex: Visual Sim-to-Real for Robust Dexterous In-hand Reorientation.” 🌐 Project page: 1/N 🧵
0:30

Advancing dexterous manipulation through scalable visual sim-to-real transfer. We are excited to share our RSS paper, “ViserDex: Visual Sim-to-Real for Robust Dexterous In-hand Reorientation.” 🌐 Project page: 1/N 🧵

Robotic Systems Lab

40,380 Aufrufe • vor 5 Monaten

We used interpretability to scale RL against open-ended tasks, cutting Gemma 12B’s hallucination rate in half by teaching it to self-correct in tandem with our probing harness.
0:29

We used interpretability to scale RL against open-ended tasks, cutting Gemma 12B’s hallucination rate in half by teaching it to self-correct in tandem with our probing harness.

Goodfire

76,490 Aufrufe • vor 8 Monaten

WUJI is showcasing its dexterous robotic hands at IROS 2026, demonstrating how human hand movements can control a robot through glove-based teleoperation. Its WUJI Hand 2 can also perform precise tasks like pen spinning using a policy trained in simulation and transferred to a real robot. These demonstrations highlight advances in robotic finger coordination, dexterity, and fine manipulation.
0:29

WUJI is showcasing its dexterous robotic hands at IROS 2026, demonstrating how human hand movements can control a robot through glove-based teleoperation. Its WUJI Hand 2 can also perform precise tasks like pen spinning using a policy trained in simulation and transferred to a real robot. These demonstrations highlight advances in robotic finger coordination, dexterity, and fine manipulation.

Space and Technology

12,949 Aufrufe • vor 26 Tagen

Unitree $6500 hand Unitree has launched a humanoid dexterous hand called the Dex5-S, featuring direct drive, backdrivability, 22 DoF, a maximum payload of 2 kg, and an aluminum-alloy design. Clearly, this hand isn’t built for heavy lifting. It’s designed for dexterous manipulation and precision tasks.
0:21

Unitree $6500 hand Unitree has launched a humanoid dexterous hand called the Dex5-S, featuring direct drive, backdrivability, 22 DoF, a maximum payload of 2 kg, and an aluminum-alloy design. Clearly, this hand isn’t built for heavy lifting. It’s designed for dexterous manipulation and precision tasks.

CyberRobo

22,623 Aufrufe • vor 1 Monat