OpenRUA

Building the open foundations for agents that act in the physical world.

Real-robot demos

What we believe

01 / 04

Embodiment is emerging in LLMs

Embodied capability is emerging in language models: within the Qwen family it jumps sharply in recent generations.

Evidence in preparation
Qwen3Qwen3.5Qwen3.6Qwen3.7Qwen3.8000031.39Embodied score (RoboDojo)Illustrative · latest point measured (uni-robot-agent)

Highlights

uni-robot-agent

A unified harness for robot use. uni-robot-agent connects open models, simulators, and real robots through shared, independently configurable tools, giving objective evaluation and generalizable agentic training a common foundation.

uni-robot-agentA unified harness for robot use.

RUAEval

Robot-use agent scores on embodied leaderboards are climbing fast. We find three loopholes that inflate them: unfair execution budgets, verifier information leaking to the agent, and skills overfit to the test tasks, and propose an evaluation contract that closes them.

RUAEvalWhy leaderboard scores overstate robot-use agent progress.