Jianglong Ye
@jianglong_ye
Ph.D. student @UCSanDiego B.E. @ZJU_China
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
Check out Dex1B! This new framework generates 1,000,000,000 diverse dexterous hand demonstrations for grasping & articulation tasks. The kicker? It uses a simple C-VAE model. 🤖✨
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
Dex1B: a groundbreaking dataset featuring 1 billion demonstrations for dexterous manipulation tasks like grasping and articulation. Built using the DexSimple generative model, it combines geometric constraints for feasibility and diverse conditions for enhanced performance.…
Huge respect for the hard work behind this — fast and accurate optimization doesn’t come easy. And with solid engineering, simple methods like CVAE can work like a charm. Congrats to the team on this amazing release! 🥳
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
We finally released Dex1B #RSS2025! Our group has been working on sim2real dexterous manipulation for the last 4 years and this is a milestone showing a grasping policy purely trained in sim with 1B data, and transfer to real world deployment without any fine tuning. The devils…
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
When it comes to scaling data, it’s not just about scale—it’s also about distribution. Leveraging generative models, even simple ones, can help improve both. Great work led by @jianglong_ye & @kaylee_keyi!
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
Dex1B is a massive dataset of 1 billion robot demonstrations for grasping & articulation tasks! Our DexSimple generative model uses geometric constraints for feasibility and conditions for diversity. Validated in sim & real-world experiments!
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
The generative model faces challenges in feasibility🔧(lower success rate compared to deterministic models) and diversity🙌(tend to interpolate rather than extrapolate). We address these by adding geometric constraints--SDF loss, and using a locally conditioned model🤖.
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
It's widely agreed that learning dexterous manipulation should leverage synthetic data, but without human priors, policies can go off the rails. Work led by @jianglong_ye demonstrates a novel approach to seed synthetic data generation—starting from human capability and pushing…
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
We have spent nearly a year from building this work to the time of a solid release. And it does accurately working towards building effective and efficient pipeline for DexHand data, which is one of the core problem for DexHand manipulation. Congrats to the team, and especially…
How to generate billion-scale manipulation demonstrations easily? Let us leverage generative models! 🤖✨ We introduce Dex1B, a framework that generates 1 BILLION diverse dexterous hand demonstrations for both grasping 🖐️and articulation 💻 tasks using a simple C-VAE model.
Introducing Co-Design for Soft Grippers via Neural Physics! 🤖 We jointly optimize gripper design and grasping poses in simulation. Neural Physics enables smoother gradients and much more efficient optimization. ⚡
For years, I’ve been tuning parameters for robot designs and controllers on specific tasks. Now we can automate this on dataset-scale. Introducing Co-Design of Soft Gripper with Neural Physics - a soft gripper trained in simulation to deform while handling load.
Check out our Human to Robot (H2R, 🧑🤝🧑➡️🤖) workshop at CoRL 2025! We aim to explore how robots can learn from diverse and rich human demonstrations. 👀
Current AI models only learn from a fraction of human intelligence. At CoRL 2025, our brand new "Human to Robot (H2R)" workshop explores how robots can learn from the vast, untapped physical human experience. sites.google.com/view/h2r-corl2… Extended abstract / paper submission deadline…
With a shared action space, EgoVLA enables robots to directly learn from diverse human videos, demonstrating strong potential for acquiring generalizable and dexterous manipulation skills ✋.
How can we leverage diverse human videos to improve robot manipulation? Excited to introduce EgoVLA — a Vision-Language-Action model trained on egocentric human videos by explicitly modeling wrist & hand motion. We build a shared action space between humans and robots, enabling…
UC San Diego Researchers Introduced Dex1B: A Billion-Scale Dataset for Dexterous Hand Manipulation in Robotics Researchers at UC San Diego have introduced Dex1B, a large-scale synthetic dataset consisting of one billion demonstrations for dexterous hand manipulation tasks,…
Test-Time Training (TTT) is now on Video! And not just a 5-second video. We can generate a full 1-min video! TTT module is an RNN module that provides an explicit and efficient memory mechanism. It models the hidden state of an RNN with a machine learning model, which is updated…
Diverse training data leads to a more robust humanoid manipulation policy, but collecting robot demonstrations is slow. Introducing our latest work, Humanoid Policy ~ Human Policy. We advocate human data as a scalable data source for co-training egocentric manipulation policy.⬇️