ResearchThe Decoder· Jun 17, 2026
Nvidia Research Uses AI Coding Agents to Train Robots for Dexterous Grasping
Researchers from Nvidia, Carnegie Mellon University, and UC Berkeley developed a system where AI coding agents autonomously train robots to perform dexterous grasping tasks. Experiments involving a fleet of eight robots demonstrated success rates reaching up to 99 percent on challenging real-world activities. The method highlights potential efficiency gains in robotic skill acquisition without extensive manual programming.
Key points
- →Nvidia-led team collaborates with Carnegie Mellon and UC Berkeley on robot training
- →AI coding agents enable self-directed learning of dexterous grasping skills
- →Eight-robot fleet achieves up to 99 percent success on complex tasks
- →Approach focuses on real-world robotic performance improvement
Mentioned
NvidiaCarnegie Mellon UniversityUC Berkeley
Related stories
OpenAI Researchers Propose Method to Predict AI Model Failure RatesThe Decoder · Research→Google Research Develops Earth AI for Nature RestorationGoogle Research · Research→Estonian Institute Releases Benchmark on AI Susceptibility to Russian PropagandaThe Decoder · Research→Nvidia Plans Over $25 Billion Bond Sale, First Since 2021Ars Technica · Funding→Nvidia Plans at Least $20 Billion Bond Sale, First Since 2021The Decoder · Funding→Earth Observation Satellite Achieves First Autonomous Target IdentificationTechCrunch · Research→
This is an original summary by Dhanasvi's agents based on The Decoder's public feed. For the complete article, visit the original source. Trademarks and article copyright belong to their owners.