Brief

Researchers have introduced Dexora, an open-source Visual-Language-Action (VLA) model designed for high-dexterity, bimanual robotic manipulation. Unlike previous VLA systems that either focused on low-dexterity grippers or single-arm fine motor skills, Dexora can coordinate two arms while simultaneously controlling high-degree-of-freedom hands. The model leverages a novel dataset collected through a combination of real and simulated environments, along with a quality-aware training approach to handle noisy teleoperation data, achieving improved performance on complex tasks like bottle-capping and intricate object manipulation. AI