JustRUA: full episode history alone substantially boosts GPT-6 robot-agent scores
Zefan Wang · hf · 2026-11-13
JustRUA is a controlled study of GPT-6 Astra robot-use agents on RoboDojo. Because reported systems differ in harness and test subsets and often change several designs at once, the team disassembles the choices in one framework.
- A lean harness that keeps full episode history and stops motions early substantially improves agent scores
- Reusing experience changes model behavior but is not a free lunch
- Scene geometry gives the largest additional benefit, even when depth is predicted from RGB alone
More from Embodied
- ASUS trains Gallop humanoid robot to pack motherboards in its own factory, targeting 99.5% success — CyberRobooo · 2026-11-13
- Foxconn demo: robot busbar assembly combining FoundationPose vision and torque feedback — lukas_m_ziegler · 2026-11-13
- Tsinghua survey reframes embodied AI as agents: L0-L5 autonomy levels — 新智元 · 2026-11-13
- NVIDIA's VLA driving model hits 20 decisions/s on Jetson Thor after Embedl optimization — No-Dragonfly6246 · 2026-11-13
- UC Berkeley's ATLAS drives humanoid robots from a single human video, no VLA or task-specific training — ChongZzZhang · 2026-11-13
- Roboticist: You Can't Do Robotics Without NVIDIA's Stack; Jensen's Long Bets Are Underrated — tianyin_xu · 2026-11-13