Humanoid robotics has entered a new phase. After decades of impressive but impractical research prototypes, multiple well-funded companies are racing to build humanoid robots that can perform useful work in real-world environments. The convergence of better actuators, foundation models for perception and planning, and massive venture investment has made 2026 a pivotal year for the field.
Tesla Optimus
Tesla's approach to humanoid robotics mirrors its approach to autonomous driving: collect massive amounts of data, train large neural networks, and iterate rapidly. Optimus Gen 2, demonstrated in late 2025, can walk at approximately 5 km/h, manipulate objects with 11-DOF hands, and perform simple assembly tasks. Tesla's advantage lies in its manufacturing infrastructure — the ability to produce actuators, batteries, and computing hardware at automotive scale and cost.
Tesla has publicly stated its goal of deploying Optimus in its own factories first, handling tasks like moving parts between stations and quality inspection. This gives the company both a captive testing ground and a clear economic benchmark: the robot must cost less per hour than the human labor it replaces.
Figure 02
Figure AI has taken a different path, partnering with OpenAI for its perception and language understanding stack. Figure 02 can hold natural conversations while performing manipulation tasks, using a vision-language model to interpret verbal instructions and visual context simultaneously. The robot demonstrated sorting, packing, and cleaning tasks in a BMW manufacturing facility pilot program.
Figure's differentiation is in human-robot interaction. Where other humanoids rely on pre-programmed task sequences, Figure 02 can accept free-form verbal instructions and break them into executable steps — a capability that dramatically reduces the programming burden for new tasks.
Agility Robotics Digit
Agility Robotics has the longest track record of any company currently building humanoid robots for commercial use. Digit, now in its third generation, is designed specifically for logistics — carrying and placing totes in warehouse environments. Agility's RoboFab facility in Salem, Oregon, has capacity to produce 10,000 units per year, making it the first humanoid robot factory at scale.
Digit's design reflects pragmatic engineering choices. It has a bird-like digitigrade leg structure that provides energy-efficient walking without the complex ankle mechanisms required by plantigrade designs. Its hands are simple but effective two-fingered grippers optimized for tote handling rather than general manipulation.
Chinese Competitors
Several Chinese companies and institutions are advancing rapidly. Unitree's H1 humanoid has demonstrated impressive athletic ability, including running at over 12 km/h. UBTECH's Walker X targets hospitality and retail applications. Fourier Intelligence's GR-1 is being developed for rehabilitation and eldercare. Government funding programs have made humanoid robotics a national priority, with significant resources flowing into actuator development, foundation models for robotic control, and manufacturing automation.
What Remains Unsolved
Despite the progress, no humanoid robot today can match human dexterity, adaptability, or endurance. Battery life is typically 2-4 hours under load. Fine manipulation — tying shoelaces, turning pages, handling flexible materials — remains extremely challenging. And reliability in unstructured, unpredictable environments is still far from industrial standards.
The race is not just to build the best robot — it is to build the first robot that is economically viable at scale. That requires solving not only the engineering but also the manufacturing, maintenance, and deployment challenges that separate a laboratory demonstration from a product.
