A team of robotics researchers has demonstrated that AI coding agents can operate entirely on their own to design, execute, and optimize training routines for robotic systems. In a recent experiment, these agents successfully instructed robotic arms to perform intricate tasks such as cutting zip ties and inserting GPUs into motherboard sockets—all without direct human supervision.
The breakthrough was made possible by ENPIRE, a newly developed agent harness framework that serves as an intermediary layer between AI models and robotic hardware. This framework equips AI agents with essential capabilities, including tool integration, contextual memory, constraint management, and feedback mechanisms. Developed by the NVIDIA GEAR (Generalist Embodied Agent Research) lab in collaboration with Carnegie Mellon University and the University of California, Berkeley, ENPIRE enables agents to autonomously plan and refine training strategies.
According to Jim Fan, NVIDIA’s director of AI, the system’s ability to self-improve has already yielded remarkable results. "A portion of our NVIDIA GEAR lab now operates tirelessly overnight," Fan noted in a LinkedIn update. "We review the outcomes each morning and track the progress."
The autonomous training process begins with an AI agent receiving a high-level task objective, such as "assemble a computer component." The agent then decomposes the task into smaller sub-tasks, writes the necessary code to control the robotic arm, and iteratively refines the process based on real-time feedback. This closed-loop system allows the agent to adjust training parameters dynamically, optimizing for speed, accuracy, or resource efficiency.
For example, when tasked with cutting zip ties, the agent first experiments with different blade angles and pressure levels before settling on an optimal approach. Similarly, when inserting GPUs into motherboard sockets, the agent adjusts its grip and alignment strategies to minimize errors. These capabilities highlight the potential of AI-driven robotic training to scale beyond pre-programmed instructions.
The implications of this research extend beyond laboratory settings. Industries reliant on automation—such as manufacturing, logistics, and electronics assembly—could benefit from systems that continuously improve their own training protocols. Traditional robot programming often requires extensive human oversight, but ENPIRE’s autonomous approach reduces the need for manual intervention, accelerating deployment timelines.
While the current system operates within controlled environments, the researchers emphasize that further refinement is needed to address real-world variability. Factors such as material inconsistencies, environmental noise, and unforeseen obstacles pose challenges that autonomous agents must learn to navigate. Nonetheless, the progress signals a shift toward more adaptive and self-sustaining robotic systems.
Looking ahead, the NVIDIA GEAR lab plans to expand ENPIRE’s capabilities, integrating additional sensors and expanding the range of tasks the system can autonomously train robots to perform. The goal is to create a generalized framework that can adapt to diverse robotic hardware and real-world scenarios, paving the way for more autonomous and intelligent automation in industries worldwide.
AI summary
NVIDIA ve Carnegie Mellon’un geliştirdiği ENPIRE adlı AI çerçevesi, robotlara kendi kendine görev öğretiyor. Yapay zeka destekli robotik eğitimi hakkında detaylar burada.
