SimEX is a framework that combines simulation and a few real robot trials to let (powered by LLMs) learn physical manipulation skills. It first has the agent iterate in simulation to build a robust toolbox of robot capabilities, then adapts that toolbox to the real world with only a handful of physical trials, using each real trial to correct the simulator and test candidate fixes. The result is that the agent can perform tasks like towel folding and barcode scanning without any demonstrations and with just 10 minutes of real robot interaction. This matters because it reduces the cost and risk of real-world robot experimentation while still achieving effective physical skills.
Simulation-first framework lets coding agents learn real robot skills in 10 minutes
SimEX autonomously builds a robot toolbox through open-ended simulation exploration, then adapts it to the real world with minimal physical trials.
Big Tech
Jiaheng Hu · Roberto Martin-Martin · Peter Stone · Rocky Duan · Zhenyu Jiang · Guanya Shi
Amazon FAR · University of Texas at Austin · Carnegie Mellon University
Research Digest··3 min read
In one line
SimEX lets coding agents learn real robot skills by first practicing in simulation and then adapting with only 10 minutes of physical trials.
Before you read this
Three ideas this paper builds onagent-driven simulationAutonomous agents automatically generate, control, or extend simulations to improve accuracy, interactivity, or performance.39 papers →coding agentAn AI-powered software system that autonomously writes, modifies, or extends source code based on high-level instructions.167 papers →agent skillsLearned capabilities that enable AI agents to perform specialized tasks with improved effectiveness.120 papers →
Read it as
The authors introduce SimEX, a two-stage autoresearch framework that tightly integrates simulation with LLM-based coding agents.
The authors designed SimEX, a framework that combines an LLM-based with a physical simulator.
How it works
stage by stageThe Daily Edition
The front page in your inbox each weekday morning: the lead, the picks, and why each one earned its place.