Figure AI, a Sunnyvale‑based robotics startup, has announced that its humanoid robots can now complete full eight‑hour work shifts autonomously. The breakthrough is powered by Helix‑02, a unified neural network system that controls walking, balance, shoulder and elbow movement, wrists, fingers, and object contact. Unlike traditional industrial systems that separate these functions into different modules, Helix‑02 learns a single model.
The robots receive data from cameras in the head and on the palms, tactile sensors on fingertips, and inertial measurement units. According to the company’s 15 May 2026 release, the platform eliminates the need for a remote operator and manual restart after each short demonstration.
Multiple robots can work side by side. In one demonstration, a humanoid unloaded and loaded a dishwasher for four consecutive minutes on a standard kitchen set‑up, recognising plates and cups, maintaining balance, adjusting grip, and avoiding the door.

Figure AI previously tested Figure 02 at BMW Group’s Spartanburg, South Carolina plant, where the robot moved over 90,000 parts and helped produce components for more than 30,000 vehicles during 10‑hour shifts.

How Helix‑02 Unifies Whole‑Body Control into a Single Neural Network

Helix‑02 is designed to solve a core problem in humanoid robotics: integrating locomotion with manipulation. In classical industrial automation, walking algorithms, arm trajectory planning, and grip force control run on separate software stacks, often written in thousands of lines of hand‑coded rules. Figure AI has replaced this with a single trainable neural network. The robot’s cameras and tactile sensors feed into Helix‑02, which then outputs commands for legs, torso, arms, and hands simultaneously.

One component, called System 0, was trained on more than 1,000 hours of human motion data. It replaces approximately 109,000 lines of hand‑written C++ code that would otherwise govern balance, gait, posture, and coordinated limb movement. Instead of rigid rules, the neural model learns to keep the robot stable while reaching, bending, or carrying objects. This approach is essential for environments where objects are not perfectly placed, people move nearby, multiple robots share a space, and item locations constantly change.

The robot’s hand control relies on tactile sensors on the fingertips. These sensors detect slip, contact force, and surface texture. When the robot unscrews a bottle cap or picks a metal part from a random pile, it must adjust grip strength in real time. Helix‑02 processes tactile feedback at millisecond speeds, allowing the robot to react to unexpected changes – for example, if a part slips or a person bumps the arm. The same network coordinates the shoulder, elbow, and wrist angles to keep the object stable during the entire manipulation sequence.

What Eight‑Hour Autonomy Means for Industrial Deployment

The claim of an eight‑hour shift without human intervention is significant for commercial viability. Previous generation humanoids – including early versions of Figure 02, Tesla’s Optimus, and Agility’s Digit – typically required frequent resets, remote teleoperation for complex tasks, or manual recovery after errors. This limited them to short demonstrations rather than real production lines. Figure AI states that Helix‑02 allows the robot to run for several minutes without a state reset or human help, and the architecture scales to full shifts by chaining multiple autonomous behaviours.

During the BMW trial, Figure 02 reportedly handled tasks such as inserting small components into assembly fixtures, inspecting parts, and transferring materials between workstations. The robot is approximately 170 cm tall, weighs 70 kg, and has a payload capacity of 20 kg.
At the Spartanburg plant, it operated alongside human workers without safety cages, using its onboard sensors to slow down or stop when people approached. The ability to work a ten‑hour shift in a live factory – moving over 90,000 parts – suggests that Helix‑02’s predecessor was already close to production readiness. The new Helix‑02 system improves on that by eliminating the need for manual restarts after each task.

For manufacturing and logistics customers, the economic argument is straightforward. A humanoid that works an eight‑hour shift with no breaks, no lunch, no overtime pay, and no complaints about working conditions can achieve a return on investment within 12 to 18 months at current hardware costs (estimated at 30,000–50,000 per unit, though Figure AI does not publicly list pricing).
The robot does not require recharging after every hour – battery life on the Figure 02 is rated for approximately five hours of continuous operation, meaning a hot‑swap battery system would be needed for a full eight‑hour shift, or the company has improved energy efficiency in the Helix‑02 configuration.
Figure AI has not released detailed power consumption numbers but claims the eight‑hour capability is validated in internal testing.

Competition with Tesla, Agility, and the Emerging Humanoid Market

Figure AI is one of several companies racing to commercialise general‑purpose humanoids. Tesla’s Optimus (now in its second generation) has been demonstrated in Tesla factories performing simple pick‑and‑place tasks, but videos show slower movement and more conservative control.
Agility Robotics’ Digit focuses on logistics – lifting totes and moving boxes – but its backward‑bending legs are less suited for factory assembly tasks that require reaching under or inside equipment. Apptronik’s Apollo is also targeting warehouse work, with a focus on safety and easy programmability.

What distinguishes Figure AI is its emphasis on end‑to‑end learning. Rather than hand‑coding individual movements, the company trains neural networks on human demonstration data.

Helix‑02 continues this philosophy: System 0 alone replaces over 100,000 lines of code. This approach allows the robot to generalise to new objects and environments without reprogramming. In the dishwasher demonstration, the robot had not been explicitly programmed for that kitchen – it recognised the plates and cups from visual training and adapted its grip based on tactile feedback.

The market for humanoids is still forming. Analysts at Goldman Sachs project a
6 billion market by 2030 and up to 154 billion by 2035 if technical challenges are solved. Current deployments remain pilot programmes. Figure AI’s partnership with BMW is one of the most advanced, alongside Amazon testing Agility’s Digit and Tesla’s internal use of Optimus. Each new video release serves as both a technical demonstration and a marketing argument to potential customers. The claim of eight‑hour autonomous shifts is precisely the kind of milestone that logistics and manufacturing executives want to see before signing contracts.

Regional Perspectives on Humanoid Deployment

In the United States, labour shortages in manufacturing and logistics are acute. The National Association of Manufacturers reported 800,000 open positions in April 2026. Humanoids like Figure AI’s are seen as a solution to fill shifts without immigration reform or wage inflation.
The Biden administration’s CHIPS Act and Inflation Reduction Act have subsidised new factory construction, but those factories cannot operate without workers. Humanoids are attractive in states like South Carolina, where BMW’s Spartanburg plant is located, as well as Texas, Ohio, and Michigan.

In the European Union, labour laws are stricter, and unions have more power. German automotive manufacturers, including BMW, Mercedes, and Volkswagen, are also testing humanoids but face worker council negotiations about task displacement. Figure AI’s eight‑hour shift claim – with no breaks – could trigger regulatory scrutiny under EU working time directives, even for robots. Some EU member states are considering “robot taxes” or mandatory human‑robot ratio laws.

In Asia, China’s UBTech and Fourier Intelligence are developing competing humanoids. Japan’s Toyota and Honda have decades of research but have been slower to commercialise. The Chinese government has set a target of 1 million humanoid robots deployed by 2030.
Figure AI’s autonomous shift capability raises the bar for all competitors – if a US startup can demonstrate eight hours of unsupervised work, state‑backed Chinese firms will accelerate their own programmes.

Figure AI’s Helix‑02 represents a genuine step forward in humanoid robotics: an eight‑hour autonomous shift without remote operation or manual resets. By unifying locomotion and manipulation into a single neural network trained on human motion, the company has reduced hand‑coded control code by more than 100,000 lines.

For industrial customers, the promise is a worker who never tires, never negotiates overtime, and never files a complaint.
For competitors like Tesla and Agility, the bar just moved. The coming year will reveal whether Helix‑02 holds up on factory floors beyond the controlled demonstration. Regardless, the era of the eight‑hour robot shift has begun.

By Jake Sullivan
May 21, 2026

Join us. Our Telegram: @forexturnkey
All to the point, no ads. A channel that doesn't tire you out, but pumps you up.