Why Dual‑Encoder Design Is Common in Humanoid Robot Joints

High‑performance servo systems and humanoid robot joints are increasingly adopting a dual‑encoder architecture, i.e., installing encoders on both the motor shaft and the output side of the reducer.

In this configuration, the motor‑side encoder handles high‑speed control and commutation reference, while the output‑side encoder provides actual end‑position feedback. This effectively compensates for position errors caused by elastic deformation and backlash in harmonic or planetary reducers. Moreover, by comparing the angular difference between the two encoders and combining it with the reducer’s stiffness model, the system can estimate joint torque, enabling collision detection and basic compliance control.

Given the space constraints of integrating frameless motors and reducers, thin, hollow‑shaft encoders that allow cabling through the centre have become the mainstream choice. Notably, the joints of Tesla Optimus, Boston Dynamics Atlas, and other renowned humanoid robots already employ this technology. Industry analysts estimate that Tesla Optimus may use more than 50 encoders per unit – 28 on rotary joints, 14 on linear joints, and 12 on the dexterous hands.