Robot perception underpins a robot’s ability to understand its environment, covering visual perception, force sensing, proprioception, and environmental modeling. Humanoid robots working in complex unstructured environments (homes, warehouses, factories) demand far more from perception systems than traditional industrial robots (fixed structured environments).
## Visual Perception Technology Stack
**RGB-D cameras (depth cameras)**: simultaneously acquiring color (RGB) and depth images — current humanoid robot mainstream visual sensors. Representative products: Intel RealSense, Azure Kinect, Orbbec (China). Depth acquisition methods: structured light (high near-range accuracy) and ToF (Time-of-Flight, for longer distances).
**Stereo cameras**: dual cameras calculating depth via disparity — precision depends on baseline distance and image resolution; no active light source needed; suited for bright outdoor environments. ZED cameras (Stereolabs) are commonly used in robotics.
**LiDAR**: high-precision 3D point cloud mapping for SLAM (Simultaneous Localization and Mapping) and environmental perception. Solid-state LiDAR (Chinese companies RoboSense, Hesai leading) costs continue falling, entering the robotics domain.
**Object recognition and 6DOF pose estimation**: robots must identify target objects’ categories and precisely estimate their 3D position and orientation (6 degrees of freedom: XYZ position + Roll/Pitch/Yaw rotation). NVIDIA FoundationPose and similar foundation models show significantly improved generalization in industrial scenarios.
## Force and Tactile Sensing
**Six-axis force-torque (F/T) sensors**: mounted at robot wrists, measuring three-directional forces (Fx, Fy, Fz) and three-directional torques (Mx, My, Mz), enabling robots to sense contact forces for compliant manipulation (screw driving, pin insertion alignment). ATI Industrial Automation is a domain leader; domestic companies (Kunwei Technology, Yuli Instruments) are catching up.
**Tactile skin**: distributed pressure sensor arrays covering robot hands and arms, providing contact location, contact force distribution, and slip detection — key for fine grasping. Current tactile skin faces durability and signal processing complexity challenges.
See [Embodied Intelligence and AI](https://sunqi.org/embodied-intelligence-ai-en/) and [Robot Joints and Actuators](https://sunqi.org/robot-actuator-motors-en/).
