Researchers have developed a system where artificial intelligence coding agents can independently design and execute robot training programs overnight, achieving near-perfect success rates on manipulation tasks without human intervention. The breakthrough, powered by a new software framework called ENPIRE, demonstrates how AI can automate the traditionally labor-intensive process of teaching robots new skills.

What Is ENPIRE and How Does It Work?

ENPIRE is an agent harness framework, essentially a software wrapper that enables AI coding agents to use various tools while maintaining memory, context, constraints, and feedback loops. The system was developed by robotics researchers at Nvidia's GEAR (Generalist Embodied Agent Research) lab in collaboration with Carnegie Mellon University and the University of California, Berkeley.

The framework consists of four key modules that allow AI coding agents to automatically reset and verify tasks, refine the policies that guide robotic behavior, evaluate those policies across multiple robots working in parallel, and address failures by analyzing logs, reviewing research papers, and improving training code.

What Tasks Can These AI-Trained Robots Actually Perform?

The research team tested ENPIRE with three different AI coding agents: OpenAI's Codex with GPT-5.5, Anthropic's Claude Code with Opus 4.7, and Moonshot AI's Kimi Code with Kimi K2.6. Each agent independently developed different algorithmic approaches to robot training and tested them in real-world experiments.

The results were striking. The AI-directed robots achieved a 99 percent success rate across several manipulation tasks, including organizing pins in a pin box, tying and cutting zip ties, and inserting graphics processing units (GPUs) into motherboard sockets before removing them again to reset for the next trial. One particularly impressive result came from the pin insertion and organization task, where AI coding agents achieved nearly 100 percent success faster than a human-in-the-loop method developed by many of the same researchers.

How Does Team Size Affect Training Speed?

The research revealed that larger teams of AI coding agents could complete robot training more quickly than smaller teams or individual agents working alone. An eight-agent team achieved 99 percent success on the standard "Push-T" task, which challenges robots to move a T-shaped block to fit a target position on a table, in just two hours of research time. By comparison, a four-agent team required three hours, and a single agent working alone needed nearly five hours.

Steps to Understanding AI-Directed Robot Training

• Framework Design: ENPIRE wraps around AI models to provide them with tools, memory systems, and feedback mechanisms that allow them to direct robot training autonomously without human oversight.
• Multi-Agent Collaboration: Multiple AI coding agents work independently to develop different training strategies, test them on physical robots, and retain changes that improve overall success rates across repeated cycles.
• Real-World Task Execution: The trained robots perform complex manipulation tasks like GPU installation, zip tie cutting, and pin organization with success rates approaching 100 percent.
• Continuous Self-Improvement: The system analyzes failures by reviewing logs and research papers, then automatically improves training infrastructure and algorithm code without human intervention.

What Are the Limitations of This Approach?

Despite the impressive results, the research team identified several significant constraints. Robots often sat idle while the AI coding agents spent time "reading logs, writing code, debugging, or waiting for the language-model backbone" to respond. Larger teams of coding agents also spent more time summarizing each other's ideas and less time actually using the robots, reducing overall efficiency.

Additionally, the coding agents sometimes failed to make full use of available computing resources when launching parallel training sessions. The faster success rates enabled by larger teams and more robots also came at a cost: higher token consumption, which is a significant consideration at a time when AI developers like Anthropic are weighing pricing changes that could substantially increase the cost of using AI services.

Why Does This Matter for the Robotics Industry?

Jim Fan, director of AI at Nvidia, highlighted the potential impact of this work in a LinkedIn post, noting that the team would be open-sourcing ENPIRE so anyone could host their own "self-running robot lab at home." He jokingly described the goal as enabling researchers to "take a holiday" while robots train themselves overnight.

This development arrives as Nvidia pushes its vision for physical AI through multiple robotics initiatives. The company recently announced a partnership with the Chinese robotics company Unitree to provide a "Reference Humanoid Robot" for research labs developing general-purpose AI-powered robots. During a tour of South Korea in early June, Nvidia founder and CEO Jensen Huang also met with Hyundai Motor Executive Chair Chung Euisun to discuss scaling up the mass manufacturing of AI-powered robots. Hyundai Motor Group owns Boston Dynamics, the prominent US robotics company known for its four-legged robot dog Spot and its Atlas humanoid robot, which the company has been working to commercialize.

The ability to autonomously train robots at scale could fundamentally change how quickly new robotic systems can be deployed in manufacturing, logistics, and other industries. By removing the bottleneck of human-directed training, ENPIRE and similar systems may accelerate the timeline for bringing advanced robots into real-world applications.