Jan Kaiser

For fifty years, industrial robot safety has meant one thing: the cage. Put the machine behind a fence, add a light curtain, cut the power if anyone gets close. That model collapses the moment a robot leaves the cage and walks on two legs through a shared workspace. Florian Weisshardt makes the case that humanoid robotics cannot inherit the safety playbook of industrial robotics, and shows how Synapticon's POSITRON platform is rewriting it from first principles:

• Safe Torque Off does not work on two legs. A humanoid is an inverted pendulum. Cutting the power doesn't make it safe, it makes it fall. Safety for legged systems requires active safe motion, not the absence of motion.
• 50+ degrees of freedom need a new safety architecture. POSITRON delivers certified (SIL3, PLe) Safe Motion monitoring across the entire kinematic chain, including safe inverse kinematics for 7-DoF arms, a scale that traditional safety PLCs cannot reach.
• From 2D LIDAR to Safe Human Detection. Perception-based safety replaces physical guards: camera-based detection with and without AI, real-time 3D person tracking. [TODO JAN: Status SIL2/PLd bestaetigen, im Mai stand "in preparation"]
• Behavioral Safety, making black-box AI certifiable. A multimodal AI layer analyzes the robot's environment and anticipates hazards before they become incidents.
• Hardware built to the standards. POSITRON STA1 (dual 800 MHz ARM) for deterministic motion safety, POSITRON STN1 (NVIDIA IGX Jetson Thor) for the AI-heavy safety stack. Both aligned to ISO 26262, IEC 61508, ISO 13849 and ISO/IEC TS 22440, certified by TUEV Rheinland.

The takeaway: humanoids will not reach real-world deployment on the back of engineering heroics and disclaimers. They will reach it on the back of a certifiable safety architecture built for how they actually move, perceive and decide. That architecture exists today.