SAN JOSE, Kalifornia (PNN) - March 15, 2026 - During a recent visit to Figure AI’s robotic company headquarters in San Jose, the author observed autonomous humanoid robots performing household, logistics and manufacturing-related tasks for extended periods without direct human intervention. According to Figure, one robot operated continuously for 67 hours with a single recorded error. The company presents this as evidence of a transition from experimental demonstrations to operational systems.
A central factor in this progress is Figure’s software architecture. The company reports that it replaced more than 100,000 lines of manually written C++ control code with a single end-to-end neural network, referred to as Helix 2. This model is designed to control the robot’s full body, including locomotion and manipulation, using learned representations rather than task-specific programming.
Instead of coding individual behaviors, Figure trains its system using large datasets derived from teleoperation and real-world interaction. Once a behavior is learned, it can be deployed across the entire robot fleet. This approach allows improvements made by one robot to be propagated to others through shared model updates.
Figure has also aligned its hardware design with this software strategy. Rather than using off-the-shelf components, the company develops its own actuators, hands, battery systems and onboard computing hardware. According to Figure, this enables faster iteration and tighter integration between hardware performance and machine learning requirements.
The company is investing in manufacturing capacity intended to scale production significantly. Figure reports that its facilities are designed to support tens of thousands of units per year, with plans to expand further. The company also plans to deploy humanoid robots within its own manufacturing processes, including assembly, testing and packaging.
Figure emphasizes that its primary performance metric is not short demonstrations, but long-duration autonomous operation in unfamiliar environments. The company defines “general robotics” as the ability for a robot to perform useful work autonomously, adapt to new settings, and operate continuously over extended periods without human oversight.
Home deployment is being approached cautiously. Figure plans limited alpha testing in residential environments in 2026, followed by broader pilot programs in subsequent years. The stated focus is on improving safety, reliability and privacy to levels significantly higher than those required in industrial settings.
Figure’s proposed business model is based on leasing robots rather than selling them outright. Under this model, customers would pay a recurring fee for access to robotic labor, similar to how human labor is compensated. Figure would retain ownership of the robots, allowing centralized software updates, monitoring and maintenance.
According to the company, demand for humanoid robots already exists in commercial and industrial sectors, with deployments underway. Figure identifies manufacturing scale and general-purpose autonomy as the primary constraints on growth, rather than market interest.
In the longer term, Figure projects that widespread deployment of humanoid robots could reduce the cost of labor-intensive goods and services. The company positions its current work as an early step toward that outcome, while acknowledging that full general-purpose autonomy remains an unsolved technical challenge.