Understanding how robots perceive and interact with everyday controls requires datasets that capture both geometry and function at a fine-grained level. Codatta’s Appliance Knobs dataset on @huggingface is designed specifically for this challenge, providing high-quality, multi-view observations of appliance knobs and rotary controls that support tasks such as 3D shape understanding, pose estimation, control-state recognition, and interaction-aware perception. Built for embodied AI, robot learning, and physical intelligence research, the dataset helps models learn the subtle visual differences that correspond to meaningful functional states in real-world devices. As part of @codatta_io broader robotics data initiative, the Appliance Knobs dataset contributes to the development of next-generation robotic foundation models capable of perceiving, reasoning about, and acting within physical environments, and is available for exploration alongside Codatta’s Manipulation Trajectory datasets through Asmora:

High-fidelity datasets are the foundation of next-generation embodied AI, robot learning, and physical intelligence and our partner @codatta_io is advancing the frontier of robotics data infrastructure on @huggingface as one of the core contributors. Its Manipulation Trajectory dataset captures fine-grained robot-object interactions with precise spatial and temporal annotations, enabling research in imitation learning, trajectory prediction, manipulation planning, and control. Complementing this, the Appliance Knobs dataset provides richly annotated multi-view observations of rotary controls to support 3D geometry understanding, state estimation, pose tracking, and interaction-aware perception. Together, these datasets help train the next generation of robotic foundation models capable of understanding and acting in the physical world. Explore Codatta's Manipulation Trajectory and Appliance Knobs datasets on Asmora today: