Robotics: Tesla's Optimus V3 production stumbles on hands and AI

Tesla's bet on humanoid robots is running into reality. The Information reported the problems, the Chinese site IT之家 picked the story up and Ars Technica analysed it. Production of Optimus V3 is hitting technical, industrial and human obstacles. Tesla has already raised output to several hundred units a week. It is aiming for more than 1,000 a week by the end of 2026.

A hundred parts in a single hand

The artificial hand remains one of the trickiest problems. To get close to the dexterity of a human hand, the hand and forearm of Optimus contain more than a hundred small parts, many of them screws. Workers still assemble these components by hand, which slows production. The equipment that positions and precisely assembles the parts is also a problem. The pace hits limits when the line speeds up. Some brand-new robots even have to be repaired by hand straight after leaving the line.

The reliability of the tactile sensors is also in question. Tesla has developed a glove-like layer of sensors. It can be replaced separately if it fails, without having to change the robot's entire hand.

AI still far from a general-purpose robot

On the intelligence side, Optimus does not yet appear to meet the requirements of a true general-purpose robot. A source cited says the robot still has to be programmed for specific tasks and operates in strictly controlled environments. That reflects a problem common to the whole sector: how to let a robot learn on its own in the real world and carry out highly varied human manipulations.

The human dimension is just as sensitive. Tesla had asked some employees at its plants in Texas and California to wear special suits that record body movements, so it could collect training data during work. The practice caused discontent among part of the workforce. Some felt they were helping train robots likely to take their jobs. The company then moved the collection to a dedicated team and set up a training centre.

These difficulties are not unique to Tesla. The company also has to face competition from other carmakers and from robotics specialists. Its founder himself acknowledged that building an autonomous humanoid robot able to take on highly diverse tasks was among the hardest problems to solve.

The open question is not whether a humanoid can walk, but whether it can learn on its own to perform varied gestures in an unprepared environment.

The business model remains unknown. The viability of the humanoid robot will depend on real, lasting and profitable deployments. How many of those there will be is still to be demonstrated.