Timeline
Optimus: a humanoid robot from prototype to production line
Four years from an AI Day slide to a converted Fremont assembly line — and still no commercial sale.
Dr. Lena Ortiz FACS
Chair, ACS Specialist Group on Autonomy and Robotics
August 2026 · 6 min read

Optimus is the clearest public example of a humanoid programme progressing faster in manufacturing readiness than in demonstrated autonomy. Both curves matter, and only one is visible.
The hardware line
Optimus was announced at Tesla's AI Day on 19 August 2021, and a walking prototype was demonstrated at the following AI Day in September 2022. Optimus Gen 2 was unveiled on 12 December 2023: lighter, with a faster walking gait, improved neck and feet, eleven degrees of freedom in the hands with tactile sensing, roughly 180cm tall and about 55kg, with a 2.3kWh pack and a vision-first perception stack derived from the FSD software line.
Reusing the vehicle autonomy stack is the programme's central engineering bet. The same occupancy and planning machinery that navigates an intersection is applied to a manipulation workspace. The bet is plausible for locomotion and coarse navigation, and considerably less obviously right for contact-rich manipulation, where force feedback rather than vision dominates.
“The constraint will not be manufacturing. It is the absent discipline for assuring learned manipulation near people.”
The production line
Tesla began converting its Fremont Model S and X production line into an Optimus assembly line during 2026; the teardown was reported completed in 46 days, with a milestone photograph posted around 1 July 2026. Reporting in July 2026 indicated Musk had asked suppliers to prepare for capacity of 100,000 units per year by the end of 2026 following Gen 3 design decisions — a capacity target, not a shipment figure.
At a meeting at Fremont reported on 20 August 2026, Tesla indicated that commercial Optimus sales are now expected to begin in 2027. Independent analyses count zero commercial units sold to date. There is no published Gen 3 specification sheet, and the long-standing sub-$20,000 unit price target has not been reaffirmed with a date.
The competence gap
If humanoid robots reach industrial deployment at anything like the announced volumes, the constraint will not be manufacturing. It will be the absence of an established discipline for assuring learned manipulation policies operating within reach of people.
Existing industrial safety standards were written for robots that are caged or that operate under speed and separation monitoring with deterministic controllers. A general-purpose learned policy fits neither model. ACS has begun scoping a competence standard for engineers who authorise such systems for shared workspaces; the Society would rather define it before a serious incident than after one.
Join the professional body behind this work
ACS members receive our research first, free CPD and ethics modules every year, and a route to professional registration assessed by their peers.
Become a memberMore from ACS Insights
Grok, Colossus and the compute arms race
xAI built a 100,000-GPU cluster in 122 days, doubled it, and merged twice. The externalities arrived with the electricity.
ArticleA national consortium to build trust in AI
ACS joins federal partners, universities and industry to strengthen assurance practice for high-impact AI systems.
Policy briefingStatute catches up: the EU AI Act, two executive orders and NIST
What actually became binding between 2023 and 2026, and what practitioners must now document.