Tesla’s optimus hits snags in hands, suppliers as scale-up begins

Tesla is ramping Optimus production from a few dozen units a week to several hundred at Fremont, repurposing the Model S/X line and aiming for an automated line that could exceed 1,000 robots weekly by year-end and ultimately 20,000. Most robots are staying inside Tesla for testing, training and controlled-task use; the V3 variant intended for customers still hasn’t met durability thresholds. Production is tripping over hands and small assemblies: each hand and forearm has over 100 screws and tiny components often put together by hand, fixtures can’t consistently align tight-tolerance parts, touch sensors have reliability problems, and some Chinese suppliers can’t hold quality at scale. Tesla plans a replaceable “sensing glove” next year to avoid swapping entire hands.

The AI can’t generalize: Optimus reportedly takes days to learn basic tasks and behaves unpredictably in untrained situations. Tesla has amassed about 500,000 hours of training data and is doubling annotation and data-collection efforts, moving self-driving staff to the robot program, deploying camera-helmet and motion-capture teams and opening training hubs. The commercial plan mirrors the FSD playbook - lease robots to similar factories to gather fleet data - while competitors like XPeng push ahead. The crucial barriers remain generalization and long-term reliability, making humanoids a niche complement to purpose-built industrial robots unless those problems are solved.