Sensing
Force, torque, tactile, depth, and environmental sensing may matter more than commodity cameras for real world manipulation.
Why it matters
A humanoid cannot work safely in the real world with vision alone. It needs to understand force, contact, balance, slip, and uncertainty.
Cameras may commoditize quickly. Force, torque, and tactile sensing are less standardized, harder to integrate, and more directly tied to manipulation quality.
Core components
- Force and torque sensors
- Tactile skins and fingertip sensors
- Depth cameras and stereo vision
- IMUs
- Proximity and safety sensors
- Sensor fusion software
Bottleneck hypothesis
Vision only approaches may cover some tasks, but manipulation and safe contact work should keep force and tactile sensing strategically important.
Signals to track
- Sensor attach rates in wrists, ankles, and hands
- Tactile skin standardization or production adoption
- Evidence that sensing improves task success, not just demo quality
- Whether camera like cost declines reach higher value sensing layers
Reference
Humanity’s Last Machine — Ch. General Sensors, Tactile Sensors, Encoders. Covers force-torque sensor architectures, tactile skin technologies (capacitive, optical, resistive taxel arrays), IMU selection for legged robots, depth cameras and LiDAR trade-offs, and encoder types (optical vs magnetic, absolute vs incremental) with resolution and reliability analysis.
Watchlist seed
Novanta/ATI, Vishay Precision, TE Connectivity, Analog Devices, Murata, TDK, Sony, Hesai, and RoboSense are diligence seeds rather than confirmed winners.