JustRUA: A Controlled Study of GPT-6 Robot-Use Agents

A controlled study, JustRUA, isolates which design choices drive gains in GPT-6 Astra robot-use agents that operate a robot arm through a fixed harness, noting that reported RoboDojo scores have risen quickly but are hard to compare because systems differ in harness and test subset and often change several designs at once. The study's stated aim is to separate the effect of individual design decisions from confounded multi-change reports.

GPT-6 Astra can drive a robot arm directly through a fixed harness, and scores of such robot-use agents on RoboDojo have risen quickly since. Reported systems differ in harness and test subset and often change several designs at once, so it is hard to tell which choices produce the gains. We take th