Tesla (robotics)

Tesla (robotics) refers to the humanoid-robot and embodied-AI program operated inside Tesla, Inc., principally the Optimus program. It is not a separate company or a separately reported business segment. Tesla describes the intended product as a general-purpose, bipedal autonomous humanoid for unsafe, repetitive, or boring tasks. Its securities filings describe Optimus as an AI robot that remains under development and commercialization.^[1][2]

This article covers the program's position inside Tesla, development history, engineering approach, public evidence, management, and manufacturing plans. Detailed descriptions of individual prototypes belong in Tesla Optimus, while claims specific to the proposed production generation belong in Optimus Gen 3.

At the research cutoff of 28 July 2026, Tesla's latest shareholder update said that first-generation Optimus production lines were being installed and that initial builds were intended for internal data collection and functionality development. The same update classified the California and Texas Optimus operations as under construction and listed no installed annual capacity.^[3] These disclosures support a pre-production or early production-preparation status. They do not establish mass production, commercial deliveries, customer revenue, or a deployed fleet.

Scope and position inside Tesla

Optimus is a program within Tesla rather than a disclosed subsidiary called "Tesla Robotics." Tesla's 2025 Form 10-K and its second-quarter 2026 Form 10-Q report only two operating and reportable segments: automotive, and energy generation and storage. Neither filing reports robotics as a separate segment or gives Optimus-specific revenue, cost, asset, or delivery figures.^[2][4] The lack of separate reporting does not mean the program is organizationally part of either reportable segment; it means investors cannot derive a stand-alone robotics business from Tesla's public segment accounts.

Tesla places Optimus within a broader effort to apply artificial intelligence to physical systems. Its AI and Robotics page groups vehicle autonomy and bipedal robotics around vision, planning, and inference hardware. The 2025 Form 10-K says that Tesla is applying AI learnings from self-driving development to robots.^[1][2] Those are company statements about a shared engineering strategy. They are not evidence that a vehicle model, vehicle training set, or complete software stack can be transferred unchanged to a humanoid robot.

The tasks differ materially. A road vehicle operates mostly on a two-dimensional surface and has a constrained set of controls. A humanoid robot must also maintain balance, coordinate many joints, make and break contacts, manipulate objects, and respond safely to people and equipment at close range. Tesla itself identifies balance, navigation, perception, and physical-world interaction as separate engineering problems for Optimus.^[1]

Development history

Concept announcement in 2021

Tesla announced "Tesla Bot" at AI Day on 19 August 2021. The presentation described a proposed humanoid that would draw on Tesla's AI, sensing, compute, battery, and actuator work. No working robot appeared: the stage demonstration used a person in a costume, and the hardware specifications and schedule presented at the event were targets.^[5] Reuters later described that appearance as the project's announcement stunt rather than a prototype demonstration.^[6]

The announcement established an intended direction, not a verified capability. In particular, the presentation's proposed dimensions, carrying capacity, speed, price, and production schedule should not be read as specifications of a completed robot. Tesla changed the hardware during subsequent prototyping, and the company has not announced a commercial product sheet that fixes those values.^[5][7]

First public prototypes in 2022

At AI Day on 30 September 2022, Tesla showed two different machines. An experimental platform later called Bumble C walked onto the stage, waved, and appeared in video performing simple handling tasks. A more integrated Tesla-built Optimus prototype was brought out on a wheeled platform and waved but did not walk during the presentation.^[6][7] Treating these as one robot obscures what the event actually demonstrated.

Tesla engineers presented actuator, hand, battery, control, and simulation work at the event. Independent roboticists interviewed by IEEE Spectrum praised the speed of the prototype effort but considered the demonstrated behaviors limited relative to established humanoid research. One expert described the locomotion stack shown in 2022 as trajectory optimization with reference controllers rather than machine learning.^[8] That observation applies to the 2022 presentation. It does not establish what later Optimus software uses.

The event was useful evidence that Tesla had built functioning hardware and an integrated development team. It did not provide controlled trials, repeated-task success rates, fall frequency, intervention rates, energy-use measurements, or independent replication.

Factory trials and demonstrations in 2024

Tesla's second-quarter 2024 shareholder update said that Optimus had begun performing tasks autonomously in one Tesla facility. A later page in the same document identified battery handling as its first task.^[9] This is a direct company claim of autonomous task performance, so it should not be rewritten as though Tesla made no autonomy claim. At the same time, the update supplied no task-completion rate, duration, number of robots, intervention rate, or protocol with which to assess the breadth and reliability of that autonomy.

At Tesla's "We, Robot" event in October 2024, Optimus machines walked among guests, served drinks, danced, and spoke. Reporting by TechCrunch, citing Bloomberg sources and an analyst note, said employees remotely oversaw many guest interactions, while the robots' walking did not require external control. A robot in one attendee video also acknowledged human assistance.^[10] The correct conclusion is not that every behavior at the event was remote-controlled, nor that the event demonstrated end-to-end autonomy. It showed a mixture of robot autonomy and human-assisted interaction whose boundaries were not fully disclosed to attendees.

Tesla's fourth-quarter 2024 shareholder update said the company had made progress on the hand, locomotion, and additional tasks and planned pilot production in 2025.^[11] "Planned pilot production" was forward-looking. The statement did not report a completed production run, output count, customer delivery, or revenue.

Management change in 2025

Milan Kovac, who had led Optimus since 2022, announced in June 2025 that he would leave Tesla. Reuters reported that Bloomberg said Ashok Elluswamy, then leading Tesla's Autopilot teams, would take responsibility for Optimus. Tesla and Elluswamy did not respond to Reuters's request for comment.^[12] The succession should therefore be attributed to reporting rather than presented as a Tesla-confirmed appointment. Tesla's later filings discuss the program but do not name a current Optimus leader.

Production preparation in 2025 and 2026

Tesla's fourth-quarter 2025 update described a planned first-quarter 2026 unveiling of Gen 3, called that design the first intended for mass production, and said production was planned before the end of 2026. It also described an eventual planned annual capacity of one million robots.^[13] These were plans at the date of the report.

The first-quarter 2026 update gave more detail about intended factory scale. It said a first-generation Fremont line was designed for one million robots per year and a future second-generation Texas line for a long-term annual capacity of ten million. However, the update's capacity table listed both operations as under construction and showed a dash rather than installed Optimus capacity.^[14] The design figures therefore describe intended line scale, not installed capacity, output, orders, or demand.

The second-quarter 2026 update is the later controlling disclosure. Tesla said it had begun construction in Fremont after decommissioning the Model S and Model X lines, that first-generation Optimus lines were being installed, and that initial builds would support an internal "Optimus Academy" for data collection and functionality development. It also said construction of the Texas building was in progress. The table continued to list both locations as under construction with no installed annual capacity.^[3]

The sequence matters because Tesla's production language changed over time:

Scroll sideways for more →

Tesla's 2025 Form 10-K said that it had not yet commercialized its robots and could not predict commercial or consumer demand.^[2] The second-quarter 2026 Form 10-Q continued to describe development, commercialization work, and investment in large-scale production. It did not disclose Optimus sales, deliveries, or revenue.^[4] Statements that Tesla was already selling Optimus, had produced thousands of commercial units, or had a verified multi-million-unit order book were unsupported by the reviewed filings.

Engineering program

Hardware

Optimus requires a mobile biped, arms and hands, sensors, onboard computing, power storage, actuators, and safety systems. Tesla's 2022 presentation showed custom actuator analysis, an integrated battery system, crash simulation, and hand development.^[7] Later shareholder updates refer to continued hand, locomotion, and task work.^[11][13] Prototype measurements and component counts have changed across generations, so generation-specific numbers should not be generalized to the whole program.

Tesla's 2025 Form 10-K identifies component development and sourcing as a material challenge. It says the program's success depends in part on Tesla's ability to source or custom-develop necessary technology and components and to make the result useful and cost-effective relative to alternatives.^[2] That filing supports a vertical-integration strategy and a supply-chain risk. It does not support a claim that every component is Tesla-designed or manufactured in-house.

Perception, planning, and control

Tesla says the target system requires computer vision, balance, navigation, perception, and interaction with the physical world. Its broader AI page discusses neural networks, planning under uncertainty, evaluation infrastructure, and custom inference hardware.^[1] The public page describes Tesla's overall AI organization, so every detail on it should not automatically be assigned to Optimus.

For a humanoid, perception must be coupled to state estimation and contact-aware control. A manipulation policy needs to locate objects and people, estimate relevant geometry, choose actions, and respond to slips, obstructions, or unexpected contact. Locomotion must coordinate balance with motion planning and manipulation rather than treat the upper body as a passive payload. Tesla's public materials describe the intended functions but do not publish an Optimus system architecture, model card, training recipe, or benchmark suite sufficient to reconstruct the current stack.^[1][3][4]

The 2022 expert assessment and the later company language should be kept separate. The former describes the control approach visible at one early demonstration.^[8] Tesla's current filings call Optimus autonomous and connect it to real-world AI data, but they do not specify which behaviors use learned policies, conventional control, scripted sequences, or remote assistance.^[2][4]

Data collection, learning, and teleoperation

Tesla's second-quarter 2026 update says that initial factory builds are intended for an Optimus Academy that will collect data and develop functionality.^[3] The document does not give the academy's dataset size, task taxonomy, labeling method, operator interface, data-release policy, or performance results. Any numerical description of the academy beyond the company's disclosed manufacturing status would be speculative.

Teleoperation can serve several distinct purposes: direct task execution, recovery from failure, safety supervision, and collection of demonstrations from which a policy can learn. Its use is not itself evidence that a robot lacks learned autonomy. The OmniH2O research system, for example, uses whole-body teleoperation both as a control interface and as a source of demonstrations for autonomous policy learning.^[15] That paper does not test Optimus, but it explains why a remotely assisted demonstration and an autonomous evaluation answer different questions.

The October 2024 event illustrates the need for behavior-level labeling. Reports supported autonomous walking but human oversight of many conversational and manipulation interactions.^[10] A precise evaluation would state, for each behavior, whether it was autonomous, scripted, teleoperated, supervised with intervention, or replayed. Without that information, a polished event video cannot establish an end-to-end autonomous system.

Compute and shared infrastructure

Tesla says that it is expanding the Cortex training clusters at Gigafactory Texas to support AI products and services, including work related to autonomous robots.^[4] This supports the existence of shared compute infrastructure. Tesla does not disclose what share of Cortex use, capital expenditure, training time, or model parameters belongs to Optimus. Company-wide AI spending should therefore not be presented as Optimus spending.

The same boundary applies to inference chips. Tesla groups custom inference hardware with its AI work and has separate chip-development programs, including Tesla AI5.^[1][14] The reviewed sources do not provide a final commercial Optimus compute specification. A chip roadmap or vehicle hardware specification is not automatically an Optimus bill of materials.

Evaluating the evidence

Public evidence about the program comes from sources with different evidentiary weight:

• Live prototype demonstrations establish that particular hardware and behaviors were shown at a particular event. They do not establish reliability outside the demonstrated conditions.
• Tesla facility and shareholder reports establish what Tesla reported, such as a battery-handling task or line construction. Unless accompanied by a protocol or third-party audit, they remain company-reported results.^[3][9]
• Edited videos can document hardware appearance and selected behaviors, but do not reveal failed attempts, operator assistance, speed changes, or selection criteria. IEEE Spectrum has specifically cautioned that a humanoid task video may be teleoperated.^[8]
• Forward-looking capacity and schedule statements document company plans. Tesla's own filings warn that such plans face new-product, manufacturing-technology, component, labor, and external uncertainties.^[4]
• Independent reporting can clarify event conditions or management changes, but it should retain attribution where Tesla has not confirmed the information.^[10][12]

Academic benchmarks also show why one successful demonstration is not enough. HumanoidBench includes whole-body locomotion and manipulation tasks and reported that contemporary reinforcement learning methods struggled with many of them; hierarchical methods performed better when supplied with robust lower-level policies.^[16] It is a simulated research benchmark, not an Optimus test. Its relevance is methodological: a credible performance claim should name tasks, initial conditions, success criteria, trial counts, baselines, and failures.

Useful public metrics for Optimus would include:

• autonomous task success over repeated trials;
• intervention and recovery rates;
• mean time between task-stopping failures;
• fall and near-fall frequency;
• grasp success across object classes and poses;
• cycle time compared with the existing process;
• energy use and operating duration under a defined workload;
• force, speed, and stopping behavior around people;
• performance after environment, lighting, object, or layout changes; and
• manufacturing yield, installed capacity, actual output, and field-support requirements.

Tesla had not supplied this set of metrics in the reviewed public filings and shareholder decks by the cutoff. That absence does not show that internal measurements do not exist. It limits what an external reader can verify.

Safety and deployment questions

Factory deployment involves risks beyond whether a robot can complete a nominal task. The US Occupational Safety and Health Administration notes that many robot accidents occur during non-routine conditions such as programming, maintenance, testing, setup, or adjustment, when workers may enter a robot's operating envelope.^[17] A humanoid that moves between work areas, handles objects, and works near people adds changing contacts and mobile hazards to that problem.

ISO 10218-1:2025 specifies safety requirements for industrial robots as partly completed machinery, while ISO 10218-2 addresses integration into complete robot applications. The Part 1 scope excludes service robots and consumer products accessible to the public.^[18] Because Tesla has discussed both factory and broader uses, the applicable safety framework would depend on the actual product, environment, and system integration. The reviewed sources do not establish Optimus certification to ISO 10218 or any other named robot-safety standard.

Important deployment evidence would include a documented risk assessment, safe operating modes, emergency-stop behavior, power and force limiting where applicable, protective separation, cybersecurity controls, maintenance lockout procedures, and validation of foreseeable misuse. Tesla's second-quarter 2026 filing says the company is exploring scalable solutions optimized for safety as well as cost, functionality, and efficiency, but it does not publish an Optimus safety case.^[4]

Commercial status and uncertainty

Tesla's filings characterize robotics as a nascent industry and identify several dependencies: applying AI training experience to robots, sourcing or developing components, cost-effectiveness, utility, competition, and the ability to scale new manufacturing technologies.^[2][4] These are the company's disclosed risks, not proof that any one risk will prevent commercialization.

The public record at the cutoff supports four conclusions:

1. Tesla has operated a real robotics program since 2021 and has repeatedly shown functioning humanoid prototypes.^[5][6][7]
2. Tesla reported a limited autonomous factory task in 2024, while other public interactions have included human assistance.^[9][10]
3. Tesla was installing first-generation production lines and constructing Optimus facilities in 2026, with initial builds assigned to internal development.^[3]
4. The reviewed filings did not report commercial Optimus deliveries, revenue, installed annual capacity, production volume, a binding price, or independently audited performance.^[2][3][4]

Those conclusions leave the program between research prototyping and industrialization. Future changes should be recorded only when a dated source distinguishes a goal from a completed milestone. Claims about exact unit counts, valuation contribution, customer orders, household capabilities, production cost, retail price, or generation-specific hardware require their own primary evidence and should not be inferred from executive forecasts.

See also

• Robot learning
• Deep learning

References

1. ^Tesla, "AI & Robotics," accessed 28 July 2026. tesla.com/AI
2. ^Tesla, "Annual Report on Form 10-K for the year ended December 31, 2025," filed 29 January 2026. sec.gov/...tsla-20251231
3. ^Tesla, "Q2 2026 Update," 22 July 2026, pp. 3, 6, and 10. assets-ir.tesla.com/...TSLA-Q2-2026-Update.pdf
4. ^Tesla, "Quarterly Report on Form 10-Q for the quarter ended June 30, 2026," filed 23 July 2026. sec.gov/...tsla-20260630
5. ^Tesla, "Tesla AI Day," 19 August 2021. youtube.com/watch
6. ^Hyunjoo Jin, "Tesla's robot waves but can't walk, yet. Musk plans to make millions of them," Reuters, 1 October 2022. investing.com/...umanoid-robot-after-delay-2902854
7. ^Tesla, "Tesla AI Day 2022," 30 September 2022. youtube.com/watch
8. ^Evan Ackerman and Erico Guizzo, "What Robotics Experts Think of Tesla's Optimus Robot," IEEE Spectrum, 4 October 2022, updated 6 April 2024. spectrum.ieee.org/...experts-on-optimus
9. ^Tesla, "Q2 2024 Update," 23 July 2024, pp. 3 and 8. digitalassets.tesla.com/...TSLA-Q2-2024-Update.pdf
10. ^Rebecca Bellan, "Tesla Optimus bots were controlled by humans during the 'We, Robot' event," TechCrunch, 14 October 2024. techcrunch.com/...humans-during-the-we-robot-event
11. ^Tesla, "Q4 and FY 2024 Update," 29 January 2025, p. 10. digitalassets.tesla.com/...TSLA-Q4-2024-Update.pdf
12. ^"Tesla's head of Optimus humanoid robot program to leave firm," Reuters, 6 June 2025. investing.com/...rm-bloomberg-news-reports-4085733
13. ^Tesla, "Q4 and FY 2025 Update," 28 January 2026, p. 8. assets-ir.tesla.com/...TSLA-Q4-2025-Update.pdf
14. ^Tesla, "Q1 2026 Update," 22 April 2026, pp. 6 and 8. assets-ir.tesla.com/...TSLA-Q1-2026-Update.pdf
15. ^Tairan He et al., "OmniH2O: Universal and Dexterous Human-to-Humanoid Whole-Body Teleoperation and Learning," arXiv:2406.08858, 13 June 2024. arxiv.org/...2406.08858
16. ^Carmelo Sferrazza et al., "HumanoidBench: Simulated Humanoid Benchmark for Whole-Body Locomotion and Manipulation," arXiv:2403.10506, revised 18 June 2024. arxiv.org/...2403.10506
17. ^US Occupational Safety and Health Administration, "Robotics: Overview," accessed 28 July 2026. osha.gov/robotics
18. ^International Organization for Standardization, "ISO 10218-1:2025, Robotics - Safety requirements - Part 1: Industrial robots," February 2025. iso.org/...73933