Clemens Eppner

@NVIDIA is working on one of the hardest problems in Physical AI so you don’t have to: generalist robotic pick-and-place. We are excited to introduce GraspGenX at #CVPR2026—a foundation model for robotic grasping that works out of the box for unknown robots, novel objects, and unseen environments. Unlike Vision-Language-Action (VLA) models or dedicated grasp networks that require expensive, embodiment-specific training, GraspGenX is cross-embodiment and works zero-shot. You simply pass a "robot prompt" alongside an image of the object to generate actions. 🚀 Key Highlights: 1) Scaling: Trained on over 2 Billion 6-DoF grasp rollouts entirely in physics simulation—a dataset size practically impossible to collect via real-world teleoperation. 2) Zero-Shot Transfer: Works out of the box for several common robot grippers widely used across the research community and industry. 3) Built for the Agentic Era: Features native MCP support, client-server architecture, and skills.md, allowing seamless integration into LLM/Agentic robotics workflows. 4) Full Pipeline Integration: Pair it with other open foundation models (like SAM3) and advanced motion solvers like cuRoboV2 for full deployment in entirely unknown environments. If you are currently executing pick-and-place with a VLA or WAM, you can use GraspGenX to generate sim-verified trajectory data and inject it into your pipeline. No need to waste precious real-world engineering hours on data collection for standard manipulation tasks. 🌐Website: graspgenx.github.io 💻Code: github.com/NVlabs/GraspGe… 📄Paper: arxiv.org/abs/2606.00998 📍CVPR Booth: Poster 619 on Jun 6 1:45 session at ExHall F This work was led by the incredible @BeiningH (Princeton), in collaboration with a phenomenal team at NVIDIA: @erwincoumans, @yu_wei_chao, @balakumar_, @clembow, and Stan Birchfield #CVPR2026

In robotics benchmarks are rarely shared. New eval setups are created for each new project, a stark difference from evals in broader ML. But generalist policies share a problem statement: do any task in any environment. Can generalist capabilities make robot evaluation easier?

Excited to share our latest work on industrial pick-and-place in collaboration with @IntrinsicAI announced at #Automate2024 Technical blog: developer.nvidia.com/blog/automatin… w/ @clembow @balakumar_ Xuning Yang

@yoavgo What's a common misconception about iid?

meh, while DL folks like to say "iid" a lot, i don't think people really understand what it means, and no DL work relies on it in any meaningful way (except maybe for some "DL theory" works, but there are kinda useless regardless).

At @NVIDIAAI research, we have been working on general-purpose robotic rearrangement 🤖 Today, we are announcing CabiNet, our recent work on scaling object rearrangement in clutter with synthetic data: cabinet-object-rearrangement.github.io w/ @a__mousavian, @clembow, @fishbotics

CabiNet: Scaling Neural Collision Detection for Object Rearrangement with Procedural Scene Generation abs: arxiv.org/abs/2304.09302 project page: cabinet-object-rearrangement.github.io

Excited to release Motion Policy Networks! w/ @fishbotics @clembow If you are at @corl_conf in NZ, checkout Adam’s presentation on Sunday, Dec 18, 9:35AM (Session 5)

Robotics AI researchers🤖 have you wondered - are we really making progress in robotics? Will we have our own ImageNet moment? Are we using the right benchmarks, if any? Join us for our Benchmarking workshop on Thursday, Dec 15 at @corl_conf sites.google.com/view/corl22ben… 🧵👇

Annoyed that existing motion generation systems promise the world but leave you with endless parameter tuning? Announcing Motion Policy Networks (MπNets)! 🤖🦾🪩   Website (code, data): mpinets.github.io Paper: arxiv.org/abs/2210.12209   W/ @Adithya_Murali_, @clembow, et. al

Motion Policy Networks abs: arxiv.org/abs/2210.12209 project page: mpinets.github.io

Very excited to share our new review paper on deep learning in grasping, focusing on 6-DoF grasping works. Got to work with some great collaborators including @clembow @a__mousavian @Juxi @leto__jean We hope the community can find our review useful! arxiv.org/abs/2207.02556

@fhuszar A little late to the party, Reviewer #2 at the chrome web store was complaining. But here you go: chrome.google.com/webstore/detai…

Quick, somebody please build a chrome extension that prepends "As part of DeepMind's mission to solve intelliegence, " to the last sentence of any abstract on arXiv.

...2019 workshop: "Pervasively neural-dynamic robotics: Do insights from neuroscience, cognitive science, and neuromorphic engineering lead to a radically new vision? (0.5d)" Tune in on Tuesday 10AM PDT to find out the title! Or click here: sites.google.com/view/rss-ai-ma… #RSS2021

Extremely electrified to be part of organizing the workshop with the second-longest title in recent #RSS history (see red dot in Fig. 1). We are only surpassed by the infamous...

We have a great lineup of speakers for our #RSS2021 workshop on understanding gaps between academic and industrial applications in #Robotics. Consider submitting extended abstracts. Deadline is tomorrow. sites.google.com/view/rss-ai-ma…

aibestpape.rs now also shows the most cited publications for each listed conference (based on Google Scholar data; only venues that are at least 6 months old).

Cool: aibestpape.rs

@syhw @ylecun The number is closer to 1100 deaths per year. (see "The Private and External Costs of Germany's Nuclear Phase-Out" by Jarvis et al., 2020. nber.org/papers/w26598)

Reminder that Fukushima is (still) causing 3690 death *per year* "in Germany". (24.6 death/TWh [1] for coal and 150TWh [2]) [1] ourworldindata.org/safest-sources… [2] en.wikipedia.org/wiki/Energy_in…

Motion planning methods need 3D model of the scene/voxelization to check for collisions. Inspired by implicit methods, we propose a model that can check 500K collision queries of unknown object against the scene in one fwd pass. Project Link: tinyurl.com/yx9we46n

ACRONYM: A Large-Scale Grasp Dataset Based on Simulation arxiv.org/abs/2011.09584