Sensors & Perception

A humanoid robot needs continuous feedback to balance, manipulate objects, detect contact, and operate safely around people. That feedback comes from several sensor classes: joint position encoders, force and torque sensors, tactile arrays, inertial measurement units, cameras, depth sensors, lidar, radar, and the processing stack that combines their signals. Performance depends on accuracy and latency, but also on calibration stability, robustness, package size, power consumption, and cost.

The most interesting bottlenecks often sit between a raw sensor and a deployable subsystem. High-quality force-torque measurement requires mechanical design, strain elements, signal conditioning, and calibration expertise. Tactile skin must survive repeated contact while routing dense signals across curved surfaces. Vision systems need suitable optics, illumination, image sensors, processors, and safety behavior. A supplier with qualified components, proprietary manufacturing, and integration knowledge may be harder to replace than a headline specification suggests.

This brief maps the sensing stack and presents five representative suppliers. The complete directory distinguishes public equities from private context companies and records theme-specific relevance, scarcity, evidence quality, thesis, and risk. It is designed to support further research into design wins and content per robot, not to treat every automotive or industrial sensor vendor as a direct humanoid beneficiary.