Verify again right here all through the week to study the newest on bodily AI, which permits machines to understand, plan and act with larger autonomy and intelligence in real-world environments.
This Nationwide Robotics Week, working by way of April 12, NVIDIA is highlighting the pioneering applied sciences which can be shaping the way forward for clever machines and driving progress throughout manufacturing, healthcare, logistics and extra.
Developments in robotics simulation and robotic studying are driving this basic shift within the trade. Plus, the emergence of world basis fashions is accelerating the evolution of AI-enabled robots able to adapting to dynamic and complicated eventualities.
For instance, by offering robotic basis fashions like NVIDIA GR00T N1, frameworks corresponding to NVIDIA Isaac Sim and Isaac Lab for robotic simulation and coaching, and artificial information technology pipelines to assist prepare robots for various duties, the NVIDIA Isaac and GR00T platforms are empowering researchers and builders to push the boundaries of robotics.
What does it take to show robots advanced decision-making in the actual world? For Nicklas Hansen, a doctoral candidate at UC San Diego and an NVIDIA Graduate Analysis Fellow, the reply lies in scalable, sturdy machine studying algorithms.
With expertise from the College of California, Berkeley, Meta AI (FAIR) and the Technical College of Denmark, Hansen is pushing the boundaries of how robots understand, plan and act in dynamic environments. Their analysis sits on the intersection of robotics, reinforcement studying and laptop imaginative and prescient — bridging the hole between simulation and real-world deployment.
Hansen’s latest work tackles one in all robotics’ hardest challenges: long-horizon manipulation. Their paper, Multi-Stage Manipulation With Demonstration-Augmented Reward, Coverage and World Mannequin Studying, introduces a framework that enhances information effectivity in sparse-reward environments through the use of multistage activity constructions.
One other key undertaking of Hansen’s, Hierarchical World Fashions as Visible Entire-Physique Humanoid Controllers, advances management methods for humanoid robots, enabling extra adaptive and humanlike actions.
Past their very own analysis, Hansen advocates for making AI-driven robotics extra accessible.
“My recommendation to anybody seeking to get began with AI for robotics is to easily mess around with the numerous open-source instruments out there and progressively begin contributing to initiatives that align together with your targets and pursuits,” they stated. “With the provision of free simulation instruments like MuJoCo, NVIDIA Isaac Lab and ManiSkill, you can also make a profound impression on the sector with out proudly owning an actual robotic.”
Hansen is the lead creator of TD-MPC2, a model-based reinforcement studying algorithm able to studying quite a lot of management duties with none area information. The algorithm is open supply and will be run on a single consumer-grade GPU.
Be taught extra about Hansen and different NVIDIA Graduate Fellowship recipients driving innovation in AI and robotics. Watch a replay of the “Graduate Program Quick Ahead” session from the NVIDIA GTC AI convention, the place doctoral college students within the NVIDIA Graduate Fellowship showcased their groundbreaking analysis.
The Seeed Studio Embodied AI Hackathon, which came about final month, introduced collectively the robotics neighborhood to showcase revolutionary initiatives utilizing the LeRobot SO-100ARM motor package.
The occasion highlighted how robotic studying is advancing AI-driven robotics, with groups efficiently integrating the NVIDIA Isaac GR00T N1 mannequin to hurry humanoid robotic improvement. A notable undertaking concerned creating leader-follower robotic pairs able to studying pick-and-place duties by post-training robotic basis fashions on real-world demonstration information.
How the undertaking labored:
- Actual-World Imitation Studying: Robots observe and mimic human-led demonstrations, recorded by way of Arducam imaginative and prescient programs and an exterior digital camera.
- Publish-Coaching Pipeline: Captured information is structured right into a modality.json dataset for environment friendly GPU-based coaching with GR00T N1.
- Bimanual Manipulation: The mannequin is optimized for controlling two robotic arms concurrently, enhancing cooperative expertise.
The dataset is now publicly out there on Hugging Face, with implementation particulars on GitHub.
Be taught extra concerning the undertaking.
The IEEE Robotics and Automation Society in March introduced the recipients of its 2025 Early Educational Profession Award, recognizing excellent contributions to the fields of robotics and automation.
This yr’s honorees — together with NVIDIA’s Shuran Tune, Abhishek Gupta and Yuke Zhu — are pioneering developments in scalable robotic studying, real-world reinforcement studying and embodied AI. Their work is shaping the following technology of clever programs, driving innovation that impacts each analysis and real-world functions.
Be taught extra concerning the award winners:
- Shuran Tune, principal analysis scientist at NVIDIA, was acknowledged for her contributions to scalable robotic studying. Notable latest papers embody:
- Abhishek Gupta, visiting professor at NVIDIA, was honored for his pioneering work in real-world robotic reinforcement studying. Notable latest papers embody:
- Yuke Zhu, principal analysis scientist at NVIDIA, was awarded for his contributions to embodied AI and broadly used open-source software program platforms. Notable latest papers embody:
These researchers will likely be acknowledged on the Worldwide Convention on Robotics and Automation in Might.
Keep updated on NVIDIA’s main robotics analysis by way of the Robotics Analysis and Improvement Digest (R2D2) tech weblog sequence, subscribing to this e-newsletter and following NVIDIA Robotics on YouTube, Discord and developer boards.
