CtrlK
| Tesla Robotics | |
|---|---|
| General information | |
| Parent company | Tesla, Inc. |
| Headquarters | Austin, Texas, United States |
| CEO | Elon Musk |
| Robotics program founded | 2021 |
| Current Optimus lead | Ashok Elluswamy (VP of AI Software) |
| Key robot | Tesla Optimus |
| Generations | Gen 1 (2022), Gen 2 (2023), Gen 3 (2026) |
| Factory sites | Fremont, California; Giga Texas, Austin |
| Website | tesla.com/AI |
Tesla, Inc. is an American electric vehicle and clean energy company that has become one of the most prominent entrants into the humanoid robot industry through its Optimus program. Announced at Tesla AI Day in August 2021, the Optimus initiative seeks to develop a general-purpose humanoid robot that leverages Tesla's expertise in autonomous driving, computer vision, neural networks, and high-volume manufacturing. CEO Elon Musk has repeatedly described Optimus as potentially "more significant than the vehicle business over time" and has projected that the robot program could generate over $10 trillion in revenue, making it the most valuable segment of the company.
As of April 2026, Tesla has deployed over 1,000 Optimus units across its factories, commenced mass production of Optimus Gen 3 components at its Fremont plant, and announced plans to convert former Model S and Model X assembly lines into a dedicated robot manufacturing facility targeting one million units per year. The company's robotics strategy is distinguished by its emphasis on vertical integration, massive data collection from factory-deployed robots, and direct technology transfer from its Full Self-Driving (FSD) platform.
Tesla's entry into humanoid robotics was formally announced on August 19, 2021, during the company's first AI Day event. Musk took the stage and unveiled the concept for the "Tesla Bot," later renamed Optimus. The company did not have a working prototype at the time; instead, a dancer dressed in a spandex robot costume performed on stage to represent the concept. The initial specification called for a robot standing 173 cm (5 ft 8 in) tall, weighing 57 kg (125 lb), with a carrying capacity of 20 kg (45 lb) and a deadlift capacity of 68 kg (150 lb).[1]
Musk framed the project as a natural extension of Tesla's existing AI capabilities. "Our cars are semi-sentient robots on wheels," Musk said during the presentation. "It kind of makes sense to put that on to a human-like form as well." The core thesis was that the same neural network architectures, training pipelines, and custom silicon that Tesla had developed for autonomous driving could be adapted to control a bipedal robot capable of navigating unstructured environments and performing physical tasks.[2]
Musk's vision for Optimus has grown progressively more ambitious over the years. At AI Day 2022, he suggested the robot could eventually cost "less than $20,000" and predicted it would be "the most significant product development that Tesla has ever done." By the Q4 2024 earnings call in January 2025, Musk stated that "Optimus has the potential to be north of $10 trillion in revenue," which would make the robot program alone larger than the combined market capitalizations of Apple, Nvidia, Microsoft, Amazon, and Alphabet. In September 2025, he posted on X that "roughly 80% of Tesla's value will be Optimus." At the 2026 Abundance Summit in March 2026, Musk reiterated his prediction that humanoid robots would outnumber humans globally by 2040, envisioning a fleet of over 10 billion units performing labor across every sector of the economy.[3][4][5]
Skeptics have challenged these projections. Rodney Brooks, co-founder of iRobot and professor emeritus at MIT, called the vision of humanoid robots as general-purpose assistants "pure fantasy thinking" in February 2026, arguing that decades of robotics research have not produced affordable, reliable manipulation systems for machines with thousands of components.[6]
The first functional Optimus prototype, nicknamed "Bumble-C," was demonstrated at Tesla's AI Day 2022 on September 30, 2022. The robot walked slowly across the stage without a tether and waved to the audience. Built primarily from off-the-shelf components, Bumble-C was rudimentary compared to competitors like Boston Dynamics' Atlas. A second, more refined prototype with Tesla-designed actuators and a custom battery pack was also shown at the event, though it could not yet walk independently and was wheeled on stage. Video footage showed this version performing simple tasks such as carrying boxes and watering plants in a Tesla factory setting.[7]
Despite the modest demonstration, Musk emphasized the rapid pace of development: the Bumble-C prototype had been built in approximately six months. The event also showcased the technology transfer from Tesla's vehicle division, including the use of Tesla's FSD computer as the robot's brain, vision-only perception (cameras rather than LiDAR), and Tesla-designed actuators adapted from vehicle components.
Tesla unveiled the second-generation Optimus in December 2023, representing a substantial leap over the Bumble-C prototype. Gen 2 featured a redesigned body that was 10 kg lighter (approximately 57 kg total), 30% faster walking speed (roughly 8 km/h), a 2-DOF articulated neck, and 11-DOF tactile hands with sensors on all fingers. The robot demonstrated the ability to pick up and transfer a raw egg without cracking it, perform a squatting motion, and walk with significantly improved balance and fluidity.[8]
Gen 2 ran on a modified version of the Tesla FSD computer and used a vision-only perception stack derived from the autonomous driving pipeline. It carried a 2.3 kWh lithium battery in its torso, designed for a full day of light-to-moderate operational tasks.
The third generation of Optimus, designated Optimus Gen 3, is Tesla's first variant designed explicitly for mass manufacturing. Key upgrades include 22-degree-of-freedom hands (doubled from Gen 2's 11 DOF), integration with Grok for natural language interaction, an OLED face display, and the Tesla AI5 chip. The Gen 3 hand system uses a tendon-based architecture with 25 actuators relocated to the forearm, providing near-human dexterity. The hand system was confirmed production-ready on February 17, 2026.[9]
Mass production of Gen 3 components began at the Fremont factory on January 21, 2026, with full-robot assembly at scale expected to begin in Summer 2026. The full Gen 3 robot is walking autonomously in Tesla offices as of April 2026, though its formal public unveiling has been delayed past the originally planned Q1 2026 window.
| Specification | Bumble-C / Gen 1 (2022) | Gen 2 (2023) | Gen 3 (2026) |
|---|---|---|---|
| Height | 173 cm (5 ft 8 in) | 180 cm (5 ft 11 in) | 173 cm (5 ft 8 in) |
| Weight | ~73 kg (161 lb) | ~57 kg (125 lb) | ~57 kg (125 lb) |
| Body DOF | 28 | 28 | 28 |
| Hand DOF | Limited (non-articulated) | 11 per hand | 22 per hand |
| Walking speed | Slow (shuffling gait) | ~8 km/h | ~8.4 km/h |
| Battery | 2.3 kWh | 2.3 kWh | 2.3 kWh |
| AI chip | Tesla FSD computer | Tesla FSD computer | Tesla AI5 |
| Voice / LLM | None | None | Grok integration |
| Target price | < $20,000 (concept) | $20,000 to $30,000 | < $20,000 at scale |
The Optimus program's technical foundation rests heavily on Tesla's Full Self-Driving autonomous driving platform. Tesla views FSD and Optimus as two applications of the same underlying AI stack, with shared perception models, planning architectures, and training infrastructure.
Like Tesla's vehicles, Optimus uses a camera-only perception system without LiDAR, radar, or other active sensors. The robot relies on a suite of cameras to perceive its environment, processing raw visual inputs through convolutional neural networks to build spatial representations of its surroundings. This approach directly mirrors the philosophy behind Tesla's FSD system, which famously removed radar and ultrasonic sensors from its vehicles starting in 2022, relying entirely on cameras and neural networks.[10]
Tesla's FSD version 12, released in early 2024, marked a fundamental shift from rule-based programming to end-to-end learning. The system replaced approximately 300,000 lines of hand-coded C++ with neural networks that learn driving behavior from millions of examples of human driving data. This same end-to-end approach has been adapted for Optimus, where the robot learns manipulation, navigation, and task execution from demonstrations rather than explicit programming. The neural network takes in raw sensor inputs and directly outputs motor commands, eliminating the need for hand-coded intermediate representations.[11]
Tesla designs its own AI chips for both vehicles and robots. The current Optimus Gen 3 runs on the AI5 chip, which delivers approximately five times the memory bandwidth of the previous-generation FSD computer. The AI5 chip uses the same neural network architecture as the FSD platform, enabling models trained for driving to share components with models trained for robotic manipulation and navigation. Tesla is also developing the AI6 chip, designed for both onboard inference and large-scale AI training, which is expected to power future Optimus generations and training clusters.[12]
Tesla's fleet of millions of vehicles equipped with cameras and FSD hardware constitutes one of the largest real-world data collection networks in the world. This "data flywheel" concept extends to Optimus: robots deployed in Tesla factories collect manipulation, navigation, and interaction data that is used to train improved neural network models, which are then pushed back to the robot fleet through over-the-air updates. The cycle of deployment, data collection, training, and redeployment mirrors the same loop that has driven improvements in FSD over the years.
Tesla Dojo was a series of custom supercomputers designed and built by Tesla for computer vision video processing and AI training. Originally conceived to accelerate FSD development, Dojo was also positioned as part of the training infrastructure for Optimus.
Tesla announced the Dojo D1 chip at AI Day 2021. Each D1 chip contained 354 custom-designed training nodes, and 25 D1 chips formed a "training tile." Multiple tiles were connected into an "ExaPod" capable of over one exaflop of compute. The first Dojo system came online at Giga Texas in July 2023.[13]
However, in August 2025, Bloomberg reported that Tesla had disbanded the Dojo team. Musk confirmed the shutdown shortly afterward, calling Dojo "an evolutionary dead end" on X. The reversal came just weeks after Tesla had projected that Dojo 2 would reach scale by 2026. Musk stated that "all paths converged to AI6" and that "Dojo 3 arguably lives on in the form of a large number of AI6 systems-on-a-chip on a single board." Tesla shifted its large-scale training workloads to clusters of Nvidia GPUs (H100 and H200), as well as planned deployments of chips from AMD and Samsung.[14][15]
Following the Dojo shutdown, Tesla's primary AI training infrastructure became the Cortex supercomputer cluster at Giga Texas. By mid-2025, Cortex had expanded to include approximately 67,000 H100-equivalent GPUs, with an additional 16,000 H200 GPUs added in Q2 2025. The cluster is used for training both FSD and Optimus models. Tesla has planned Cortex to reach 250 MW of power capacity by approximately April 2026, with full 500 MW capacity by mid-2026.[16]
Tesla's strategy for advancing the Optimus program centers on deploying robots internally at its own manufacturing facilities before offering them to external customers. This approach serves two purposes: collecting real-world operational data to improve the AI, and validating reliability and safety in controlled industrial settings.
| Year | Milestone |
|---|---|
| 2024 | First Optimus units placed in Tesla factories for testing; initial tasks include battery cell sorting |
| Early 2025 | Musk targets 5,000 units for internal factory use by end of 2025 |
| Mid-2025 | Actual production falls well short of target; only several hundred units produced |
| April 2025 | China's rare earth export restrictions disrupt magnet supply chain |
| January 2026 | Gen 3 component mass production begins at Fremont; over 1,000 total units deployed across factories |
| Q1 2026 | Robots primarily in "R&D and learning phase" rather than performing productive work |
Tasks performed by factory-deployed Optimus units have included battery cell sorting and handling, parts transport between stations, quality inspection, and basic pick-and-place operations. However, Musk acknowledged during Tesla's Q4 2025 earnings call in January 2026 that no robots were doing "useful work" yet, characterizing the factory deployments as primarily for learning and data collection.[17]
In April 2025, China blocked the export of seven rare earth metals in response to U.S. tariffs on Chinese goods. The restrictions affected neodymium-iron-boron (NdFeB) magnets, of which each Optimus unit requires approximately 3.5 kg. Since China supplies around 90% of the world's rare earth magnets, the restrictions created a significant bottleneck for Tesla's production plans. Musk confirmed the impact during a public statement, saying that Tesla was working with Chinese authorities to secure an export license and was simultaneously pursuing alternative suppliers in Africa, Australia, and the United States.[18]
On January 28, 2026, during Tesla's Q4 2025 earnings call, Musk announced that Tesla would discontinue production of the Model S and Model X at the end of Q2 2026. The assembly lines at the Fremont, California factory that had produced these vehicles since 2012 and 2015, respectively, would be converted to manufacture Optimus robots. The decision was significant: while the Model S and Model X were Tesla's original flagship vehicles, they accounted for less than 3% of total deliveries in 2025, with the Model Y outselling them by a factor of roughly 20 to 1. Tesla targets a production run rate of one million Optimus units per year from the converted Fremont lines.[19]
In late 2025, Tesla broke ground on a dedicated Optimus manufacturing facility at Gigafactory Texas with an ambitious target capacity of 10 million units per year. This facility is planned as the primary high-volume production site, with mass production targeted for 2027. The scale of the planned facility reflects Musk's vision of eventually producing tens of millions, and ultimately 100 million, Optimus units annually as future generations (Optimus 4, 5, and beyond) are developed.[20]
| Year | Target | Actual / Status |
|---|---|---|
| 2024 | ~10,000 units (internal plan); "several thousand" (Musk) | Low hundreds; testing and prototyping phase |
| 2025 | 5,000 units for internal use; parts for 10,000-12,000 | Several hundred produced; < 10% of target |
| 2026 | 50,000 units (original target) | Gen 3 component production started January 2026; full assembly expected Summer 2026 |
| 2027 | 1 million/year (Fremont); Giga Texas factory online | Planned |
| 2027+ | 10 million/year capacity (Giga Texas) | Planned |
Tesla's production targets have consistently outpaced actual output. The 2025 goal of 5,000 internal units was missed by a wide margin, with manufacturing totals reaching only several hundred to roughly 1,000 units by early 2026. Contributing factors included the rare earth supply chain disruption, the leadership transition following Milan Kovac's departure, and the decision to redesign certain Gen 3 components before resuming full-scale production.[21]
Elon Musk serves as Tesla's CEO and has been the primary public champion of the Optimus program. Musk personally announced the project at AI Day 2021 and has set the production targets, pricing goals, and long-term vision for the robot. His involvement ranges from high-level strategy to specific technical decisions, such as the commitment to vision-only perception and end-to-end neural networks.
Milan Kovac joined Tesla in April 2016 as a Staff Software Engineer on the Autopilot team after working at SoftKinetic, a Belgian company specializing in 3D gesture recognition that was acquired by Sony in 2015. He was quickly promoted to Engineering Manager of Autopilot Software, and from 2019 to 2022 he oversaw the development of Tesla's second-generation Autopilot system. He transitioned to lead the Optimus initiative and was eventually promoted to Senior Vice President and Head of Engineering for the Optimus program.
On June 6, 2025, Kovac announced his departure from Tesla, citing a desire to spend more time with his family. He posted on X that the decision was "the most difficult decision" of his life, adding that his "support for Elon Musk and the team is ironclad." His departure came at a critical juncture for the program and contributed to production delays, as the team underwent a leadership transition and paused to address hardware design issues under new direction.[22][23]
In January 2026, Hyundai Motor Group appointed Kovac as a group adviser and outside director for Boston Dynamics, placing him at Tesla's most technically advanced competitor in the humanoid robotics space.[24]
Ashok Elluswamy was the first software engineer hired for Tesla's Autopilot team, joining the company in June 2014. Born in Chennai, India, he earned a Bachelor of Engineering in Electronics and Communication from the College of Engineering, Guindy, and a Master of Science in Robotic Systems Development from Carnegie Mellon University. Before Tesla, he worked at WABCO Vehicle Control Systems and as a research intern at the Volkswagen Electronic Research Lab.[25]
At Tesla, Elluswamy rose from senior engineer to Director of Autopilot Software (May 2019) and eventually to Vice President of Autopilot and AI Software. Following Kovac's departure in June 2025, Elluswamy assumed leadership of the Optimus program, consolidating oversight of both FSD and the humanoid robot under a single executive. This organizational change tightened the integration between Tesla's autonomous driving and robotics efforts, reflecting the company's view that both programs share the same foundational AI stack.[26]
Konstantinos Laskaris serves as the Lead Director of the Optimus program, responsible for day-to-day technical leadership. In April 2026, Laskaris gave a keynote presentation at the ETH Robotics Club in Zurich, where he shared details about Gen 3 development and compared Tesla's developmental velocity to a "Formula One car that increases its acceleration in proportion to its speed." He has also led recruitment efforts focused on gear manufacturing and electrical integration to address the challenges of high-volume robot production.[27]
In June 2025, Tesla filed a lawsuit against Zhongjie "Jay" Li, a former Optimus engineer, and his startup Proception Inc. in the U.S. District Court for the Northern District of California. Tesla alleged that Li downloaded Optimus-related files onto personal devices in the weeks before his departure in September 2024, and that Proception, which Li and co-founder Jack Xu (another former Tesla engineer) incorporated less than a week after Li left the company, had "successfully built" advanced humanoid robotic hands that bore a "striking resemblance" to designs Li worked on at Tesla. The lawsuit highlighted the intensity of competition for humanoid robotics talent and intellectual property.[28]
Musk's price targets for Optimus have evolved over time but have consistently aimed at consumer affordability:
| Date | Stated price target | Context |
|---|---|---|
| AI Day 2022 (September 2022) | Less than $20,000 | Initial concept target |
| "We, Robot" event (October 2024) | ~$30,000 at scale | Revised estimate |
| Q4 2024 earnings call (January 2025) | $20,000 to $25,000 | Long-term goal |
| 2026 | Below $20,000 at scale | Gen 3 target |
Current manufacturing costs are estimated at $50,000 to $100,000 per unit. Initial commercial units sold to external customers in late 2026 are expected to carry prices in the $100,000 to $150,000 range. The sub-$20,000 target depends on achieving production volumes of millions of units per year, which remains years away. For comparison, Chinese competitor Unitree Robotics launched its R1 humanoid at $5,900 in July 2025, demonstrating that aggressive pricing from Asian manufacturers could challenge Tesla's value proposition even before it reaches scale.[29]
On October 11, 2024, Tesla held a high-profile event titled "We, Robot" at Warner Bros. Studios in Los Angeles. The event featured dozens of Optimus robots interacting with attendees: serving drinks, playing charades, conversing, and performing dance moves. The spectacle generated enormous media attention and positioned Tesla as a leader in humanoid robotics consumer experience.
However, within days of the event, reports emerged that the Optimus robots' conversations and complex manipulations were being remotely controlled by human operators. YouTuber Marques Brownlee noted that his charades game with an Optimus was either "the single greatest robotics and large language model demonstration the world has ever seen, or it's mostly remote operated by a human." A Tesla engineer subsequently confirmed that human technicians were operating the robots from behind the scenes for all conversational and fine-manipulation tasks. The robots walked autonomously, but essentially all interactive behavior was teleoperated.[30][31]
Tesla did not initially disclose the teleoperation, leading to criticism about transparency. The incident drew comparisons to the 2021 AI Day reveal, where a dancer in a costume stood in for a nonexistent robot. Critics argued that Tesla was overstating Optimus's capabilities, while supporters countered that teleoperation is a standard development technique used to collect training data for future autonomous operation.
Tesla enters an increasingly crowded humanoid robot market that has grown rapidly since 2023. The company faces competition from both well-funded Western startups and Chinese manufacturers that have achieved higher production volumes.
| Company | Robot | Units sold/deployed (2025) | Price range | Key advantage |
|---|---|---|---|---|
| Tesla | Optimus | ~1,000 (internal) | $100K-$150K (initial) | Manufacturing scale; FSD data flywheel |
| Boston Dynamics | Atlas (electric) | Limited commercial | ~$140,000+ | Best-in-class agility; 50 kg lift capacity |
| Figure AI | Figure 02 | Pilot deployments (BMW) | $100,000+ | Vision-language model integration; BMW partnership |
| Agility Robotics | Digit | Commercial deployments | Enterprise pricing | Purpose-built for logistics; Amazon deployments |
| Unitree Robotics | G1, H1 | 5,500 (2025) | $5,900 (R1) | Lowest price point; highest volume |
| Agibot | Multiple models | 5,168 (2025); 10,000 by March 2026 | Competitive | Second-highest production volume globally |
| 1X Technologies | NEO | Pilot programs | TBD | Consumer-focused design |
Chinese manufacturers have emerged as formidable competitors. A 2026 report from Rest of World noted that Chinese companies controlled approximately 90% of the humanoid robot market by unit volume, with Unitree and Agibot each selling more units in 2025 than Tesla's stated production target of 5,000. However, U.S. firms, including Tesla and Figure AI, lead in AI sophistication, valuation, and investment.[32]
Tesla's primary competitive advantages are its manufacturing expertise, vertical integration (designing its own chips, actuators, and software), massive real-world data collection infrastructure, and willingness to invest billions in dedicated production capacity. Its primary disadvantages include late entry relative to established robotics firms, repeatedly missed production timelines, and the gap between Musk's public statements and actual operational capabilities.
The Optimus program has attracted sustained criticism from the robotics research community and financial analysts:
Aggressive timelines. Musk's production and capability projections have consistently proven optimistic. Every major milestone announced between 2021 and 2025 has been delivered late or significantly scaled back. Critics draw parallels to Tesla's Full Self-Driving program, which was first described as feature-complete by 2018 and remains a driver-assistance system rather than a fully autonomous one as of 2026.
Autonomy gap. As of early 2026, Tesla's own earnings calls confirm that factory-deployed Optimus units are in an R&D and learning phase rather than performing productive manufacturing tasks. The gap between demonstration capabilities (often teleoperated) and genuine autonomous operation remains significant.
Manipulation complexity. Robotics researchers have long identified dexterous manipulation as one of the hardest unsolved problems in the field. Rodney Brooks has specifically described hands as the "Achilles' heel" of humanoid robotics, arguing that the mechanical complexity, sensor requirements, and control algorithms needed for reliable general-purpose manipulation are far from solved.[6]
Humanoid form factor. Some experts question whether a humanoid form is necessary or optimal for most industrial tasks. Specialized robots, such as industrial robot arms or mobile platforms, are significantly more efficient and reliable for specific applications. The humanoid design adds complexity in balance, power consumption, and cost without clear advantages for many factory use cases.
Valuation concerns. Financial analysts have questioned whether Musk's projections of $10 trillion in Optimus revenue are realistic, particularly given the early state of the technology and the mismatch between announced targets and actual production volumes.
Supporters counter that Tesla's manufacturing prowess, massive capital reserves, and unique data collection infrastructure (from both its vehicle fleet and factory-deployed robots) give it advantages that smaller robotics companies cannot replicate. The sheer volume of real-world data Tesla can collect, combined with its custom AI training infrastructure, may accelerate progress faster than skeptics anticipate.