Berkeley Artificial Intelligence Research

伯克利人工智能研究实验室BAIRCommon

A joint AI lab at UC Berkeley, and a major hub for robot learning research.

BAIR brings together UC Berkeley research groups working on computer vision, machine learning, natural language processing, planning, and robotics, and launched the BAIR blog in 2017 to publish its results. It is one of the most prolific academic institutions in robot learning: Sergey Levine's group has produced real-world deep reinforcement learning, BridgeData V2, Octo, SERL, and HIL-SERL; Pieter Abbeel's group has been highly influential in imitation learning and reinforcement learning; Ken Goldberg's AUTOLAB built the grasping network Dex-Net; and Jitendra Malik's group developed Rapid Motor Adaptation (RMA) for legged locomotion control. A number of embodied AI companies have spun out of BAIR — Covariant, for instance, was founded by Abbeel and his students, and Levine is a co-founder of Physical Intelligence.

ExampleThe open-source generalist robot policy Octo was led by a BAIR team and trained on Open X-Embodiment data.

Also called: Berkeley AI Research
Related: Physical Intelligence · Covariant · Octo · BridgeData V2 · HIL-SERL · Dex-Net 2.0
Sources: Welcome to the BAIR Blog
Covariant (company) - Wikipedia