Stanford REAL

University lab producing influential open models, datasets and evaluation for generalist robot learning.

No.Paper · what it does and showsIndustry contribution

OpenVLA: An Open-Source Vision-Language-Action Model2024 ↗

A 7B open-weight vision-language-action model trained on about 970,000 robot episodes from Open X-Embodiment, fine-tunable to new tasks and embodiments with modest compute.

Gave industry and academia an affordable, inspectable alternative to proprietary robot models, with fine-tuning feasible on consumer GPUs.

Learning Fine-Grained Bimanual Manipulation with Low-Cost Hardware2023 ↗

Presents ALOHA, a low-cost bimanual teleoperation rig, plus Action Chunking with Transformers, which learns fine-grained tasks such as threading, slotting and opening translucent objects.

The hardware design and ACT algorithm were widely copied, letting small teams and startups collect high-quality manipulation data without six-figure robot arms.