Tesla has officially unveiled Optimus Gen 3, its first mass‑produced humanoid robot, marking a turning point in robotics and automation. Building on the prototypes shown at AI Day 2022 and 2024, Gen 3 is designed for real‑world factory and household tasks – from assembling cars on Tesla’s production lines to folding laundry, cleaning floors, and even cooking. The robot stands 5’8” (173 cm), weighs 62 kg (137 lbs), and can carry up to 20 kg. Under the hood is Tesla’s new AI 5 inference computer (5nm, 500 TOPS) running a specialized version of the Tesla FSD (Full Self‑Driving) world model adapted for physical manipulation. The robot features 24 precision actuators (12 rotary, 12 linear), each with force control and torque sensing, enabling smooth, human‑like motion. Battery life is rated for 8 hours of continuous operation (swappable pack, 2.3 kWh). The biggest news: price – starting at $19,999, with volume production ramping to 1 million units per year by 2028. Tesla claims the robot can replace repetitive labour with a payback period of under 12 months for businesses. Pre‑orders open June 1, 2026, with first deliveries in November 2026. This article covers the hardware, AI, real‑world use cases, safety, and how it compares to competitors like Boston Dynamics’ Atlas, Figure 02, and Unitree H1.
Hardware Deep Dive: Actuators, Hands, and Battery
Optimus Gen 3’s actuator design is a major differentiator. The rotary actuators in shoulders, hips, and neck use Tesla’s ‘inrunner’ motor with a custom harmonic drive (ratio 1:100). Peak torque 140 Nm at 25 rad/s. Linear actuators in elbows, knees, and ankles use a planetary roller screw (99% efficiency) driven by a brushless DC motor. The hands have 11 degrees of freedom each (3 fingers + thumb, each with two joints), with magnetic encoders for position. Fingertips have a silicone ‘grip pad’ with micro‑suckers for holding smooth objects. Battery cells are Tesla’s 4680 format (12 cells in 4S3P), providing 2.3 kWh at 52V. An onboard charger supports 2 kW fast charging – 20% to 80% in 25 minutes.
AI & Software: How Optimus Learns New Tasks
Tesla uses a 3‑stage training pipeline. Stage 1: A human wears a lightweight exoskeleton (motion capture suit) and performs a task 5‑10 times – the suit records joint angles, force, and hand pressure. Stage 2: The data is fed into a neural network (ViT + transformer) that learns the policy. Stage 3: The robot practices in simulation (NVIDIA Isaac Sim) with domain randomisation (varying friction, lighting, object positions) – 1 million simulated attempts per real demo. The final policy is compressed and deployed via OTA. Current skills include: picking and placing, sweeping, folding, screw driving, and even making coffee.
Factory vs. Home: Use Cases Unveiled
Tesla’s own factories will be the first customer: Optimus Gen 3 will work alongside humans on assembly lines – part insertion, quality inspection, and material handling. For homes, Tesla demonstrated folding laundry, loading a dishwasher, and vacuuming. The robot can follow voice commands (via integrated microphones and speaker) and navigate using a built‑in SLAM system. However, stairs remain a challenge – the current model can only climb slopes up to 15°. Tesla promises a ‘stair upgrade kit’ in 2027.
Price & Business Model: Robot‑as‑a‑Service or Buy?
Customers can buy Optimus outright for $19,999 (base) or $24,999 (pro with better hands). But Tesla also offers a leasing option: $500/month for 36 months (includes maintenance and software updates). For large fleets (100+ units), Tesla provides on‑site support and dedicated cloud training. Elon Musk claims a 12‑month payback for businesses replacing $15/hour labour in 3 shifts ($131k/year saved per robot). Critics question maintenance costs and downtime.
Safety & Regulations: Will Optimus Hurt People?
Tesla submitted Optimus to UL (Underwriters Laboratories) for safety certification – expected UL 3300 (robotic safety standard) by Q1 2027. The robot’s maximum force output is limited to 150 N in ‘home mode’ (equivalent to a 15 kg push). It cannot lift more than 20 kg, reducing crush risk. In factories, a ‘safety cage’ is required for first‑year deployment until certified. Tesla also includes a ‘nanny mode’ for homes: the robot moves at half speed and issues audible warnings when near children or pets.
Competitors: Optimus vs Figure 02 vs Atlas vs Unitree
Figure 02 (priced $30k, expected 2026) has similar specs but slower inference (200 TOPS). Boston Dynamics’ Atlas (research only, no price) is more agile but hydraulically actuated (noisy, leak risk). Unitree H1 ($90k) is fast but lacks fine manipulation. Optimus stands out on price and Tesla’s manufacturing scale – but critics point out that no humanoid has yet proven reliable for 8‑hour shifts. Early third‑party teardowns will be crucial.
Should You Pre‑order? A Buyer’s Guide
For businesses with repetitive material handling, pre‑order makes sense – refundable deposit is only $100. For home use, wait for Gen 3.5 or later; the current model cannot climb stairs and struggles with very cluttered environments. Tech enthusiasts may enjoy it as a development platform – Tesla will release a developer SDK in 2027. We recommend waiting for independent reviews and safety certifications before deploying near people.
Key Highlights
AI 5 Inference Computer (500 TOPS)
Custom 5nm neural accelerator with 32 GB memory, dedicated motor control processor, and Tesla’s FSD world model adapted for humanoid locomotion and manipulation.
24 Precision Actuators (Force‑Controlled)
12 rotary + 12 linear actuators, each with torque sensing and backdrivability. Peak torque 140 Nm, force up to 4,500 N. Enables smooth, human‑like motion.
8‑Hour Swappable Battery
2.3 kWh lithium‑ion pack located in the chest. Hot‑swappable in 30 seconds. Battery management system predicts remaining runtime with 95% accuracy.
End‑to‑End Neural Manipulation
No hand‑coding – robot learns tasks from video demonstrations using imitation learning and reinforcement learning in simulation. Generalises across object positions and lighting.
Tactile Sensor Array (1,400 points)
Pressure sensors in palms, fingers, and feet. Can detect slip, texture, and contact force down to 0.1 N. Enables delicate tasks like handling eggs or glassware.
360° Camera Vision + Depth Sensing
8 autostereo cameras with overlapping fields of view. Each camera provides RGB + monocular depth estimation. Works in low light (0.1 lux).
Cloud Learning & Fleet Update
Robots can share anonymised learning data (opt‑in) to improve the global model. Tesla pushes over‑the‑air updates every 2 weeks – new skills, improved balance, etc.
Safety Features: Soft Skin + E‑Stop
Impact‑absorbing foam under the outer shell reduces injury risk. Torque cutoff within 2 ms of unexpected contact. Remote emergency stop via Bluetooth.
Pros
- ✓Unbeatable price for a full humanoid robot ($19,999)
- ✓High payload (20 kg) and decent battery life (8 hours)
- ✓End‑to‑end neural learning reduces programming effort
- ✓Swappable battery allows 24/7 operation with spare packs
- ✓Over‑the‑air updates continuously improve skills
- ✓Backdrivable actuators enable safe human interaction
- ✓Tactile sensing allows delicate object handling
- ✓Large ecosystem (Tesla factories, fleet learning) gives data advantage
- ✓Leasing option lowers entry barrier
Cons
- ✗Cannot climb stairs (only ramps up to 15°)
- ✗Limited to 20 kg lift – less than a human worker in many industries
- ✗No certification for home use yet (expected late 2027)
- ✗8‑hour battery may be insufficient for 3‑shift operation without hot‑swap
- ✗Software still lacks generalisation – new objects require retraining
- ✗Potential safety risks if software fails (e.g., torque cutoff malfunction)
- ✗High upfront cost still prohibitive for individual consumers
- ✗Repair and maintenance not yet priced or widely available