EmbodiedEvo lets coding agents build reusable robot skills in simulation

Santiago (@svpino) says a coding agent such as the Codex CLI can control a simulated robot by writing and running Python code, check its results, and save reusable procedures as skills.

Santiago says those skills can transfer to a real robot in a different environment without retraining a model. He links to EmbodiedEvo, a project from Einsia AI that publishes the method, in which coding agents accumulate skills and lessons in a simulated playground. The project reports a 45.08 average score and 38.95% success, and ranks first on the RoboDojo leaderboard. These figures come from the project as cited in Santiago's post, not from an independent evaluation.