Data for Physical AI and Robotics includes multimodal datasets (image, video, LiDAR, audio, sensor data) used to train systems in perception, decision-making, and action. This supports a wide range of systems such as autonomous vehicles, drones, humanoid robots, industrial robots, cobots, and Autonomous Mobile Robots (AMRs) operating in smart factory environments.

Physical AI systems rely on diverse data types, including 3D LiDAR cuboid annotation, sensor fusion data, egocentric datasets, and motion capture (MOCAP) data. These datasets often integrate inputs from sensors such as cameras, IMU (Inertial Measurement Units), and IR markers, as well as markerless tracking systems, to support Vision-Language-Action (VLA) models and vision AI agents.

Robotics and embodied AI systems operate in real-world environments, requiring multimodal data to understand context, combine perception signals, and execute actions reliably through sensor fusion and real-time processing.

Trajectory collection captures sequences of movements for navigation and control, while teleoperation records human-guided actions. These datasets often include kinematic data (joint positions, velocities) and are essential for training robots in manipulation, navigation, and real-world interaction.

Grasp point labeling identifies where and how a robot should interact with objects. It is critical for manipulation tasks in robotics, especially for industrial robots, cobots, and humanoid systems performing fine-grained interaction.

High-quality annotated data enables accurate perception, including object detection, tracking, and environment understanding. This is essential for applications like Autonomous Mobile Robots (AMRs), drones, cobots, and smart factory systems.

Motion capture (MOCAP) records human movement, including body, hand, and pose data. It can be collected using marker-based systems (e.g., IR markers) or markerless approaches, enabling robots to replicate human actions and improve interaction in embodied AI systems.

In smart factory environments, data supports visual inspection, workflow automation, and robotic interaction across industrial robots, cobots, and AMRs. Data factory pipelines ensure continuous data collection, annotation, and optimization.

VLA models combine visual, language, and action data to enable robots to understand instructions and perform tasks. They rely on multimodal datasets, including image-text pairs, trajectory data, kinematic signals, and interaction annotations.

Edge AI allows data processing directly on devices, enabling faster response times and real-time decision-making. This is particularly important for systems like AMRs, drones, and cobots that rely on onboard sensors such as cameras and IMUs for low-latency operation.