Talk to Synthra

Perception, reasoning, action. One loop.

The Synthra intelligence platform is the embodied-AI system that runs every U1 robot. Multimodal perception, vision-language-action reasoning, spatial understanding, navigation, and control work as one continuous loop, so the robot understands a live restaurant and decides its own actions instead of replaying programmed routes.

See the U1 Series
Illustrative film: one U1 through one service: it learns the room, takes an order, decides a route, adapts, delivers, docks, then scales to a fleet.
Illustrative

One service with U1

Each chapter links to the section below that explains it.

  1. It enters the restaurant and learns the room.

    Deployment

  2. It takes the order in the guest’s own words.

    Conversation

  3. It decides its path from the room and the task.

    The loop

  4. When the room changes, it decides a new route.

    No fixed paths

  5. It stops at the right table, within the guest’s reach.

    Venue model

  6. It returns to its dock and charges on its own.

    Docking

  7. On U1 Max, robots share the floor as one fleet.

    Fleets

How does a Synthra robot decide what to do?

It runs a continuous loop. It sees the room through multimodal perception, understands what it sees with vision-language intelligence, decides an action from the task and the situation, acts through navigation and control, and adapts as people and objects move. Then the loop runs again, for as long as the robot is working.

Illustrative still: a U1’s scan sweeps a dining room as points resolve tables, chairs, seated guests and a standing server.
Illustrative

Close-up of Decide · One service with U1

Follow one task through the loop: a pasta and a sparkling water for table 04, while a server crosses the aisle.

Illustrative still: a U1’s scan sweeps a dining room as points resolve tables, chairs, seated guests and a standing server.
Illustrative

See

Multimodal perception

The robot perceives people, tables, objects, pathways, and service areas, and notices what has changed since the moment before. Perception is continuous, not a scan taken once at setup.

Takes in
The scene around it, what people say to it, and its own motion.
Produces
People, furniture, free space, and changes, located in the room.
At table 04
A guest is seated at table 04. A server is walking out from the kitchen.
Illustrative still: the room is labelled DOCK, KITCHEN, CUSTOMER, SERVER, TABLE 04 and PATHWAY, with free floor marked in the aisle.
Illustrative

Understand

Vision-language intelligence

Detections become meaning. Vision-language intelligence connects what the robot sees with what it knows about the venue and the task: which table is the destination, who is a guest and who is staff, what each area is for.

Takes in
Perceived entities, the venue model, the menu, and the current order.
Produces
Context: who is where, what each place is for, and what is happening now.
At table 04
Table 04 is the destination, and the seated guest is the customer. The server is staff, carrying plates into the aisle.
Illustrative still: candidate routes fan out toward TABLE 04, blocked ones stop short, and one is decided in blue as the service path.
Illustrative

Decide

Contextual reasoning

The robot chooses what to do next from three things: the environment, the task, and the situation. That covers where to go, which way to approach, when to wait, and which task comes first.

Takes in
Context, the active task, and the restaurant’s operating parameters.
Produces
An action and a path, decided now, not retrieved from a stored route.
At table 04
Take the service path along the open aisle to table 04.
Illustrative still: the U1 moves along its service path as a server, now labelled MOVING OBSTACLE, walks toward the aisle.
Illustrative

Act

Navigation and control

Navigation and control turn the decision into physical movement: following the path, keeping clear of people, carrying the order, and stopping where the guest can reach it.

Takes in
The decided action and path.
Produces
Motion through the room, arrival, and delivery at the table.
At table 04
The robot follows the service path toward table 04. The server steps into it, and the path is blocked.
Illustrative still: a server stops in the aisle, the route ahead turns amber as BLOCKED, and an alternate path re-forms to TABLE 04.
Illustrative

Adapt

Continuous update

Every result feeds back into perception. When a guest stands, a path closes, or a new request arrives, the robot updates its understanding and its plan without abandoning the task.

Takes in
What changed while it acted.
Produces
A revised plan, and the next pass through the loop.
At table 04
An alternate path around the server is decided. The robot reaches table 04 and stops with the order within the guest’s reach.

It sees the restaurant change. Then changes with it.

No predefined route.

A conventional restaurant robot follows a route configured in advance, waypoint to waypoint, so when a person steps into the aisle it holds until the route clears. A Synthra robot plans its path at the moment of service from what it perceives, and when a person, a chair, or a cart gets in the way, it decides another one.

Close-up of Adapt · One service with U1

Illustrative still: a preprogrammed route runs through waypoints A, B and C while a person stands beside a table, off the route.
Illustrative

Conventional robot

  1. Preprogrammed route
  2. Waypoint A
  3. Waypoint B
  4. Waypoint C
  5. Stop
Before service
Route stored in advance: waypoint A, B, C, stop.
A person steps into the aisle
The stored route runs straight through them.
What happens next
It holds at the person. Its route was fixed before service.

Synthra robot

  1. Perceive
  2. Understand
  3. Navigate
  4. Adapt
Before service
No stored route. It perceives the room and knows the destination.
A person steps into the aisle
The person is perceived and understood as a moving obstacle.
What happens next
It decides a new route around them. The delivery continues.
Route-following compared with contextual navigation
Conventional route-followingSynthra contextual navigation
Where the route comes fromConfigured by hand before service, waypoint by waypointDecided by the robot at the moment of service
What the robot knowsCoordinates and waypointsTables, people, service areas, and free space
When the aisle is blockedIt waits on its fixed routeIt perceives the obstacle and plans a new path
When the room changesThe route stays the sameThe plan changes with it
What setup involvesRoutes and workflows programmed in advanceThe robot explores the venue and builds its own spatial context
Read the open questions in autonomous navigation research

What does a Synthra robot understand about a restaurant?

Everything that shapes service, held in one working model of the venue. Places: tables, kitchen and service areas, pathways, destinations, and docks. People and movement: guests, staff, moving people, obstacles, traffic, and other Synthra robots. Service: orders, menu items, and the context around each task. The model updates while the robot works.

Venue modelIllustrative top-down model of a restaurant floor in layers. Places: eight tables, the kitchen pass, a bar, a buffet, open pathways, a standby area and a dock. People and movement: guests, servers in motion, a cart in an aisle, a busy zone at the pass, and two other Synthra robots. Service: an order at the pass linked to table 04 with a decided path.TABLE 03TABLE 02TABLE 01TABLE 08TABLE 07TABLE 06TABLE 05BARBUFFETKITCHEN · PASSSTANDBY AREADOCKAVAILABLE PATHWAYSTABLE 04CHANGING TRAFFICCUSTOMERCUSTOMERSERVERSERVEROBSTACLESYNTHRA ROBOTSYNTHRA ROBOTORDER · PASTA, SPARKLING WATERSERVICE TASKDESTINATION
Model
Venue model
Loop
Understand
Status
Illustrative
Figure
Top-down plan
Venue model. One model, three layers of meaning. Switch layers to see each. Illustrative figure, not measured data.

Places

Tables
Destinations with a name. Table 04 is a place with seats and guests, not a coordinate.
Kitchen areas
Where finished orders are collected, and where staff traffic is heaviest.
Service areas
Bars, stations, and buffets, where the robot works alongside staff.
Available pathways
The free space that exists right now, not corridors drawn in advance.
Destinations
Where the current task ends: a table, the pass, the standby area, or the dock.
Docking locations
Where the robot charges, designated by the restaurant.

People and movement

Customers
Guests seated, standing, arriving, or asking for something.
Restaurant staff
Servers and kitchen staff, whose work the robot plans around.
Moving people
Anyone in motion, whose path the robot keeps clear of.
Obstacles
Chairs pushed back, bags on the floor, carts, anything new in the way.
Changing traffic
How busy each part of the floor is, and how that shifts during service.
Other Synthra robots
Robots on the same floor, understood as robots with tasks, not as obstacles.

Service

Orders
What was asked for, by whom, and for which table.
Menu items
Dishes and drinks, with the descriptions and dietary information the restaurant uploads.
Service context
The situation around each task: who asked, what has been served, what should happen next.

The restaurant sets the standby area and docking locations and uploads the menu. The robot builds the rest of the model itself and keeps it current during service.

Spatial intelligence in Synthra research

How is a Synthra robot deployed in a restaurant?

In six steps, and most of them belong to the robot. It enters the venue, explores it, and builds its own understanding of tables, service areas, pathways, and docks. The restaurant configures its menu, service information, and operating parameters. Then the robot serves, and docks itself when it is done.

No fixed workflow. Deployment is built on understanding the venue, not on programming every step.

Close-up of Arrive · One service with U1

Illustrative still: a U1 glides in through the lit doors of a dark venue it has never seen, its blue core lit.
Illustrative
  1. Robot
    The robot arrives in a venue it has never seen.
  2. Robot
    It moves through the space and perceives tables, pathways, service areas, and obstacles.
  3. Robot
    It builds spatial context: what each area is for, where guests sit, where service happens.
  4. Restaurant
    The restaurant uploads its menu, menu descriptions, dietary and service information, and operating parameters, and sets the standby area and docking location.
  5. Robot
    It talks with guests, takes orders, delivers food and drinks to the right table, and adapts as the room changes.
  6. Robot
    Between tasks it returns to standby, or docks and charges on its own.

Configuration · from the restaurant

Menu
The dishes and drinks the robot can offer and take orders for.
Menu descriptions
What the robot draws on to answer questions about a dish.
Dietary information
Ingredients and dietary notes, for questions like “vegetarian and not spicy”.
Service information
How the venue runs service, from table service to buffet.
Operating parameters
The rules the restaurant sets for how and when the robot operates.
Standby area
Where the robot waits between tasks.
Docking location
Where the robot charges.

What the restaurant does not program

  • Routes between tables
  • Waypoints
  • Table-by-table paths

The robot builds its understanding of the venue itself.

Plan a deployment

What it does once it is serving

  • Talks with guests and takes orders in their own words
  • Answers menu and dietary questions from what the restaurant uploaded
  • Works out which table each order belongs to
  • Collects food from the kitchen and delivers food and drinks
  • Navigates on its own around people and moving obstacles
  • Keeps working as the floor changes during service
  • Returns to its standby area between tasks
  • Docks and charges on its own
  • Coordinates with other Synthra robots where supported (fleets on U1 Max)

How does a conversation become a service task?

The robot answers from the menu and dietary information the restaurant uploaded, takes the order in the guest’s own words, and works out the rest itself: which table the guest is at, where to collect each item, and how to get there. The exchange becomes a structured task it then carries out.

The exchange

  1. Guest

    What’s vegetarian and not spicy?

  2. U1

    I can recommend the mushroom truffle pasta or the roasted vegetable bowl. The vegetable bowl can also be prepared without the chili dressing.

  3. Guest

    Bring me the pasta and sparkling water.

  4. U1

    One mushroom truffle pasta and a sparkling water for table four.

Order · One service with U1

Illustrative still: a guest at Table 04 orders the pasta and sparkling water, and the words become one service task.
Illustrative

Service task · Table 04

Order
Mushroom truffle pasta, sparkling water
Table context
Table 04 · the guest asked for vegetarian, not spicy
Service task
Collect the pasta at the kitchen pass and the water at the bar, deliver together
Destination
Table 04, approached from the open side
Execution
Path decided at dispatch, re-decided as the room changes
  1. Queued
  2. En route
  3. Delivered
  4. Docked
What happens after delivery

The guest never named the table. The robot knew where the question came from.

See how U1 works a full service

How does automatic docking work?

When a task ends, the robot decides what comes next: wait in the standby area for the next request, or charge. It plans a path to its designated dock through the room as it is, aligns, connects, and charges, then reports itself available again. No one has to send it back or plug it in.

Dock · One service with U1

Illustrative still: U1 on its dock, the charge ring closed and green, labelled available.
Illustrative
  1. Task complete
    Table 04 has its order. The task closes.
  2. Standby decision
    Wait for the next request, or charge now.
  3. Navigation
    A path to the dock, decided through the room as it is.
  4. Docking
    It aligns with the dock and connects.
  5. Charging
    It charges at the location the restaurant designated.
  6. Available
    Charged, it reports itself ready for the next task.

One intelligence platform. Three physical configurations.

U1e, U1, and U1 Max run the same Synthra intelligence platform. Everything described above runs on all three. What changes is the body: footprint, payload, endurance, maneuverability, and the scale of service each one is built for. U1 Max adds fleet coordination.

The same on every U1

  • Multimodal perception
  • Vision-language-action
  • Spatial understanding
  • Contextual navigation
  • Conversational service
  • Automatic docking

Different by model

  • Footprint
  • Payload
  • Endurance
  • Maneuvering space
  • Service scale
  • Deployment density
  • Fleet coordination, on U1 Max

  • U1e Compact

    Full intelligence. Smaller footprint.

  • U1 General-purpose

    The general-purpose hospitality robot.

  • U1 Max High-capacity, fleet-native

    Built to work as a fleet.

How do U1 Max robots work as one fleet?

They share one understanding of the venue and talk to each other. Each request goes to the robot best placed for it, by position, workload, and available capacity. Charging is coordinated, traffic at busy points is managed, and service requests are prioritized, so several robots behave as one service system.

Fleet, cluster, and multi-node coordination is a U1 Max capability. U1e and U1 recognize other Synthra robots on the floor.

Close-up of Fleet · One service with U1

Illustrative still: ROBOT 04 detects people in an aisle, the shared map updates across six U1 Max robots and ROBOT 02 reroutes.
Illustrative

Fleet beats

Illustrative scenario

ROBOT 04 perceives a change in the main aisle and shares it with the fleet.
RobotStatusDestinationPathCharging stateAvailable capacity
ROBOT 01Loading at Pass 2Pass 2Stopped at Pass 2ReadyLoading
ROBOT 02Rerouted by the shared map (changed)KitchenRe-decided around the change in the main aisle (changed)ReadyCarrying
ROBOT 03Standing byStandby areaStopped in the standby areaReadyAvailable
ROBOT 04Change detected and shared (changed)BarToward the bar, clear of the changeReadyAvailable
ROBOT 05En routeTable 18West along the cross-aisle to Table 18ReadyCarrying
ROBOT 06ServingTable 24Stopped at Table 24LowCarrying
  • Distributed task assignment
  • Robot-to-robot communication
  • Location awareness
  • Workload distribution
  • Charging coordination
  • Shared environmental understanding
  • Traffic management
  • Service prioritization
Explore U1 Max

Questions about the Synthra intelligence platform

Short answers for operators, researchers, and partners evaluating the platform. For anything specific to your venue, talk to Synthra. For the full record, see what Synthra has published, and what is pending.

What is embodied AI in a service robot?

Embodied AI is intelligence that is expressed through a physical body acting in the real world. In a Synthra robot, perception, language, reasoning, navigation, and control run as one loop, so understanding a request, seeing the room, and moving through it are parts of the same decision.

What is a vision-language-action model?

A vision-language-action model connects what a robot sees, what it is asked, and what it does. In a Synthra robot, a request such as “bring me the pasta” is grounded in the scene: “me” becomes the guest at table four, “the pasta” an order at the kitchen pass, and both become actions navigation and control can execute.

Can a Synthra robot tell guests from staff?

Yes. Customers and restaurant staff are separate entities in the robot’s model of the venue. A guest seated at a table is a customer and a likely destination; a server carrying plates is staff, whose path the robot plans around. The distinction shapes where the robot goes, whom it serves, and whose path it keeps clear.

How does a Synthra robot treat other robots on the floor?

As robots with tasks, not as obstacles. Every U1 model recognizes other Synthra robots on the same floor and plans around them. On U1 Max, recognition becomes coordination: the robots share one map of the venue, divide incoming requests, take turns at narrow crossings, and stagger their charging so requests stay covered.

Does changing the menu mean reprogramming the robot?

No. The menu, menu descriptions, and dietary information are content the restaurant provides, not code or routes. The robot answers questions and takes orders from that content, so changing the menu means updating what the restaurant has provided. Nothing about navigation or service routes has to be rewritten.

Are the visuals on this page real footage?

No. The sequences on this page are illustrative motion graphics that show how the Synthra intelligence platform works. They are not recordings of a specific venue, robot, or deployment. When footage of U1 robots is published, it will be labelled as footage and replace these illustrations.

Autonomy doesn’t stop when the environment changes.

See the platform in its general-purpose body, or bring your venue, your service, and your questions to Synthra.

Meet U1Talk to Synthra