Toyota has placed a large number on the possible expansion of factory automation. The automaker estimates that deploying robots across its own plants, group companies and major suppliers could require roughly 400,000 machines and annual spending of about 1 trillion yen, or $6.4 billion, beginning in 2028.
That estimate is not an approved investment plan. Toyota has not said whether it will spend the full amount or how many years expenditure at that level might continue. The figure nevertheless shows the scale at which the company is considering physical AI: a mix of industrial machines, automated logistics and future systems designed to work more flexibly around people and changing production conditions.
A Large Estimate Built on Existing Factory Work
The 400,000 total includes both replacements for current equipment and new installations, covering humanoid and non-humanoid robots. Toyota already uses robotics for parts transport and picking, and its financial reporting has described demonstrations in production and medical-equipment logistics.
One example is the piston assembly line at Toyota’s Kamigo Plant. The line began operating with robots in January 2025 and later moved from three operators to full automation, according to Toyota’s integrated report. Its history also illustrates why buying machines is only part of the challenge.
Kamigo began introducing robots on selected sub-lines in 2008. When engine production expanded overseas, some facilities lacked enough skilled maintenance workers to keep the equipment running. Toyota returned certain processes to manual production while transferring manufacturing knowledge from Japan. Workers developed jigs and tools that the company said doubled or tripled efficiency in some tasks, and that experience later informed newer automation.
The episode complicates any simple projection from robot count to productivity. Automation can remove labor from a process, but it also creates requirements for maintenance, reliable data and expertise at every site where the system operates.
Physical AI Targets Less Predictable Tasks
Toyota’s newer work focuses on machines that can respond when components are not perfectly positioned. Its KumiPro parts-picking robot uses cameras to identify loosely arranged objects and is already operating at Toyota Motor East Japan. A demonstrated assembly process adds force-feedback control so a robot can compensate when camera recognition does not exactly match an object’s physical position.
The company is also developing ELEY, short for Embodied Learning robot for Enhanced Yield. It has two human-like arms and an omnidirectional base, and is intended to continue working when physical contact produces forces or positions different from those expected by its control system.
Toyota describes this category as physical AI because sensors, decision systems and motion controls work together to perceive conditions and adjust actions. The company has identified long-duration reliability, repeatable precision and the data infrastructure for robot learning as unfinished areas. It plans to test ELEY in environments closer to production and train with both successful and failed attempts.
Those limitations matter more than a polished demonstration. A factory system must repeat useful behavior over long shifts and handle variation without turning every exception into downtime or a safety problem.
Simulation Must Survive the Factory Floor
Toyota is training humanoid robots through reinforcement learning in thousands of parallel simulations. The method can accelerate practice, but the company acknowledges a persistent simulation-to-reality gap. Sensor readings, floor friction and actuator behavior differ outside a virtual environment, so researchers vary simulated conditions and feed results from physical machines back into training.
Toyota Research Institute is separately working with Boston Dynamics on AI control for Atlas. In a 2025 demonstration, the humanoid completed walking, lifting, sorting and packing in one sequence using a single Large Behavior Model, with new skills introduced through human demonstrations.
Japan already had about 450,500 industrial robots operating in 2024, according to figures cited from the International Federation of Robotics, and installed another 44,500 that year. Its automotive industry installed roughly 13,000, an 11 percent annual increase and the highest total since 2020. Toyota’s estimate is broader than those statistics because it includes replacements and multiple robot types across an extended supplier network.
The proposed scale therefore remains a planning signal, not a deployment forecast. Toyota has shown pieces of the technical and operational path, while its own experience also shows that maintenance capacity, production reliability and knowledge transfer will determine how much of the $6.4 billion vision becomes working physical AI.
