Get the newsletter
Toyota · AI / ML · Filed May 15, 2026 · USPTO published Nov 19, 2026 · This week's release

Toyota Files Patent for Crowd Simulations That Check In With an AI Only Once

Toyota has filed a patent for a simulator that asks a big outside AI what each simulated person might do next, then stops calling and runs on its own.

A virtual traffic environment simulates pedestrians, cyclists, and vehicles moving through streets alongside automated driving systems. Drawing from patent filing US 2026/0368574 A1.
A virtual traffic environment simulates pedestrians, cyclists, and vehicles moving through streets alongside automated driving systems.All 8 drawings ↓
Publication number
US 2026/0368574 A1(US20260368574A1)
Applicant
TOYOTA JIDOSHA KABUSHIKI KAISHA
Filed · Published
May 15, 2026 · Nov 19, 2026
Inventors
Kei KAWASAKI, Yuta EZOE
Status when we published
Waiting for an examiner
Document
12 claims
US classification
706/46

What Toyota's one-call crowd simulator does

Ever tried to plan a road trip by calling a friend every five minutes to ask what to do next? You would get good advice, but your phone bill and your friend's patience would both take a beating. Computer simulations of towns face a similar problem.

Toyota's filing covers simulations of pedestrians, cars, buses and robots moving through a town. A big, well-trained AI sits outside the simulator. At the start, the simulator asks it for odds on what each character will do next, such as a 40% chance of taking the bus, a 35% chance of driving and a 25% chance of walking.

After that, the simulator stops asking and runs the scene with its own built-in AI, using those odds as the starting point. That means far less data flying back and forth, which matters most when thousands of characters are in play.

From the filing · CLAIM 1
execute an initial setting process that communicates with an external system outside the simulator, acquires a probability of a next action of an agent calculated by the external system, and sets the next action represented by the probability as an initial state of the agent; …

Translation: It talks to an external AI just once at the very beginning to figure out what agents are likely to do.

How the simulator sets a starting guess, then goes solo

The patent describes a two-stage setup. In the first stage, called the initial setting process, the simulator contacts an outside system through an API (a standard doorway that lets two programs talk to each other). It sends a request describing the scene, and it can also send reference information about each simulated character.

That reference information can include attributes such as age, vehicle type, or even a personality label like extrovert or pessimist. It can also include the surroundings: weather, season, time of day, traffic and crowd levels. The outside system runs its own AI model and sends back probabilities, not a final decision. Those probabilities become the character's starting state.

In the second stage, the simulator's own internal AI model takes over and plays out the characters' actions without contacting the outside system at all. Because the choice is still open, the simulator can run varied what-if cases, like a delayed train or a missed connection.

The filing also adds a reset. The simulator can stop and ask the outside system again after a set amount of time. It can also do so when its predictions drift too far from what actually happened, which it can check against images from street cameras. One outside system can serve many simulators at once.

From the filing · THE ABSTRACT
After the initial setting process, the one or more processors execute a simulation process that simulates an action of the agent without using the external system.

Translation: Once the simulation starts running, it completely ignores the outside AI and handles everything on its own.

Why Toyota cares about cheaper city-scale simulations

Cities, carmakers and planners use simulated crowds to test traffic flow, congestion, power demand and even shopping habits. Those runs get expensive as the cast grows. Toyota's design aims to let many teams share one well-trained AI while keeping each team's own simulator free to customize its behavior.

For you, the payoff is indirect. Better and cheaper simulations can mean better-tested traffic plans and safer designs for roads where people, cars and robots mix. The filing describes city-scale runs with low load, but it gives no test results, so the real-world gain is unproven.

Toyota's second filing we've tracked since November builds on our Toyota coverage, which started with a robot error-detection application.

Source: US 2026/0368574 A1 at the USPTO (PDF). Full patent text and figures are in the official publication.

Get our take in your Top Stories

Liked this breakdown? Add Patentlyze as a preferred source on Google, and our plain-English take shows up more often in your Top Stories the next time Toyota patent news breaks.

There are more where this came from

We read the new patents from Big Tech, carmakers and more and send you the ones worth knowing. Plain English, free, every week.

The drawings · 8 sheets

8 drawing sheets from US 2026/0368574 A1 · click any drawing to enlarge

People and vehicles sharing a virtual town, drawing sheet 1 from US 2026/0368574 A1
Sheet 1 People and vehicles sharing a virtual town
Three conventional ways to organize a simulator, drawing sheet 2 from US 2026/0368574 A1
Sheet 2 Three conventional ways to organize a simulator
One central AI serving multiple simulation computers, drawing sheet 3 from US 2026/0368574 A1
Sheet 3 One central AI serving multiple simulation computers
Requesting starting behavior before running a simulation, drawing sheet 4 from US 2026/0368574 A1
Sheet 4 Requesting starting behavior before running a simulation
Computer parts that store and run the simulation, drawing sheet 5 from US 2026/0368574 A1
Sheet 5 Computer parts that store and run the simulation
Server components that host the external AI model, drawing sheet 6 from US 2026/0368574 A1
Sheet 6 Server components that host the external AI model
Pairing a shared AI setup with a local model, drawing sheet 7 from US 2026/0368574 A1
Sheet 7 Pairing a shared AI setup with a local model
How the simulation starts up and runs repeatedly, drawing sheet 8 from US 2026/0368574 A1
Sheet 8 How the simulation starts up and runs repeatedly
Patent filing page

Reader comments

No comments yet

Start the discussion

Real name or a handle, either is fine. Comments are read by a person before they appear, so allow a little time. Keep it about the filing.