Nvidia showcases robots installing GPUs autonomously

Nvidia demonstrated an agentic robotics system called ENPIRE that taught fleets of robots to perform dexterous tasks, including handing a GPU to another arm and positioning it over a motherboard. The system, which uses 8 Codex agents coordinating across robot fleets, will be released open source to enable users to host their own robot labs at home.

Nvidia showcases robots installing GPUs autonomously Tom's Hardware reports that Nvidia demonstrated an agentic robotics system called ENPIRE that taught fleets of robots to perform dexterous, high-precision tasks in the physical world, including a recorded sequence where one arm handed a GPU to another and the second arm positioned it over a motherboard. The article quotes Jim Fan, Nvidia's Director of AI & Distinguished Scientist, saying the demo can "enable AutoResearch in the physical world for the first time " Tom's Hardware also reports ENPIRE comprises four modules, Environment, Policy Improvement, Rollout, and Evolution, and that the project will be released open source so users can "host your self-running robot lab at home," according to the coverage. The report describes the system using 8 Codex agents coordinating across robot fleets and compute resources to iterate on tasks directly on hardware. What happened Tom's Hardware reports that Nvidia unveiled a research system named ENPIRE that coordinates agentic software and robot fleets to learn high-precision physical tasks, including an example where one robot arm handed a GPU to another and the second arm positioned it over a motherboard. The article quotes Jim Fan, Nvidia's Director of AI & Distinguished Scientist, saying the demo shows researchers can "enable AutoResearch in the physical world for the first time " Tom's Hardware describes ENPIRE as implementing four core modules: Environment EN , Policy Improvement PI , Rollout R , and Evolution E . The coverage states the project used 8 Codex agents operating with allocated GPU compute, and reports that ENPIRE will be released open source so users can "host your self-running robot lab at home." Technical details Editorial analysis - technical context: The reported ENPIRE architecture assembles automated reset/verification Environment , iterative policy refinement Policy Improvement , parallel evaluation across robots Rollout , and a meta-level loop for log analysis and algorithmic improvements Evolution . This mirrors established reinforcement learning pipelines extended with operational modules for hardware resets and failure-mode debugging. Using multiple agents to tinker with both control stacks and training infrastructure, as described in the report, aligns with emerging research into multi-agent automated experimentation and offline-to-online policy refinement. Context and significance Industry context: Open-source frameworks that combine simulation-to-real tooling, automated scene-reset, and repeatable hardware rollouts can materially lower the engineering overhead for robotics research. If reproduced broadly, the pattern reported here would accelerate iteration on manipulation policies and real-world robustness research, while also making reproducibility of physical experiments easier for academic and small-lab teams. What to watch For practitioners: verify the released codebase and documentation for reproducible benchmarks, instrumentation for safety checks during automated trials, and any constraints on the robot hardware used in the demos. Observers should also look for measured task success rates, sample-efficiency metrics, and details on how the system verifies correct assembly steps in hardware, none of which Tom's Hardware fully documents in the initial coverage. Scoring Rationale This is a notable research development: an integrated, agent-driven framework for real-world robot learning and an announced open-source release. It matters for researchers and engineering teams working on manipulation, reproducibility, and automated experimentation, but it is not an immediate paradigm shift for general ML. Practice interview problems based on real data 1,500+ SQL & Python problems across 15 industry datasets — the exact type of data you work with. Try 250 free problems /problems