Back to overview
Division 02In development

Humanoid Robotics

AI software for humanoid robots in the home, care, industry and logistics.

We develop modular software that lets a humanoid robot perceive its surroundings, map them, navigate safely, plan tasks and handle objects — with people always in control.

Concept visual of a BELTH humanoid standing in a modern home interior
  • Perception
  • Mapping
  • Planning
  • Navigation
  • Manipulation
  • Safety
Engineers working on a mobile manipulator beside a simulation workstation

Introduction

From software to a genuinely useful robot assistant

A robot only becomes useful once it understands what is happening around it and picks the right sequence of actions by itself. That is exactly where our work sits: the software that turns perception into useful action.

BELTH builds that software layer and the accompanying product concepts. The robot itself comes from a manufacturer. That way we combine our knowledge of AI and autonomy with the production capacity of partners who already build hardware.

  • 01A control stack that can run on existing platforms
  • 02Simulation in ROS 2 and Gazebo before we touch hardware
  • 03Clear agreements on IP, licensing and division of roles

Software stack

Six modules, one coherent system

Each module works on its own, but only becomes powerful in combination.

  1. 01

    Perception

    The robot sees and recognises people, objects and obstacles through cameras and sensors.

  2. 02

    Mapping

    It builds an up-to-date 3D map of the space and knows where it is.

  3. 03

    Planning

    A task is translated into a logical sequence of sub-steps.

  4. 04

    Navigation

    It picks a safe route and adjusts as soon as something changes.

  5. 05

    Manipulation

    Grasping, carrying and placing objects with the right force and precision.

  6. 06

    Safety

    Limits, emergency stop and human oversight are built into every layer.

Studio concept visual of the BELTH humanoid platform

Product direction

A robot that learns how you live

Most humanoid programmes answer the question "what can this robot do?". BELTH is developing an answer to a different one: what can you teach it? A dedicated page sets out the direction — learning the home, learning from demonstration, household memory, and why none of it is allowed to touch the safety layer.

  • Learns the home as a structured world, not a floor plan
  • Taught by demonstration, not by recorded trajectories
  • Household memory that makes assistance proactive
  • An independent safety layer that cannot be trained away
Explore Learning & Personalization
Concept visual of a humanoid placing a bottle on a kitchen counter

Application areas

Where a robot assistant makes a difference

Four domains BELTH is developing this software for. None of them is a shipping product today.

Concept visual of a humanoid carrying a tray through a living room
01

Household support

Helping with tidying, carrying and recurring chores at home, so people can live independently for longer.

Concept visual of a service robot carrying supplies in a care facility corridor
02

Care and assisted living

Supporting care staff and residents: fetching items, keeping an eye out and simply being present, without taking over the care itself.

Concept visual of a wheeled humanoid platform transporting a crate indoors
03

Industry and production

Taking over repetitive operations on the line, even when the setup changes regularly.

Concept visual of a mobile manipulator picking from racking in a warehouse aisle
04

Warehousing and logistics

Assembling orders, moving boxes and feeding material between people and machines.

How it works

From perceiving to acting safely

Four steps the robot runs through again and again.

  1. 01

    Perceive

    Sensors capture the environment: who is there, what is where, what is moving.

  2. 02

    Understand

    The software links that picture to the task and works out what is needed.

  3. 03

    Act

    The robot performs the steps: moving, picking up, handing over, putting away.

  4. 04

    Supervise safely

    Every action is checked and logged, and can always be stopped by a person.

The core

One software layer, many environments

A living room, a care room, a production hall and a warehouse look completely different, yet they ask exactly the same of a robot: see what is there, understand what needs to happen and act safely. We build that layer properly once, then tune it per environment.

Concept visual of three BELTH platform variants: a humanoid, a wheeled humanoid and a mobile arm
Concept overview of six robot platform variants on a neutral background
  • 01The same modules for perception, planning and safety
  • 02A dedicated task library and behaviour rules per environment
  • 03Human oversight built in from the start, not bolted on

Development status

Where the project stands today

We are open about what is finished and what still lies ahead.

Engineers reviewing perception data while a robot moves through a marked test area
  1. 01In place

    Software architecture

    The modular structure from perception through to safety is settled.

  2. 02In progress

    Simulation in ROS 2 and Gazebo

    We test behaviour, navigation and manipulation in a virtual environment.

  3. 03Next phase

    Physical integration

    The next phase: running our software on a real robot platform.

  4. 04Partner wanted

    Manufacturer and pilot partners

    We are looking for a hardware partner and a test environment to take that step.

Divisions

Four domains, one technology base

The same perception, navigation and integration modules return in every division. What runs in software today is the foundation for what we are building in robotics and autonomous mobility.

Collaboration

Four ways to work together

Every track starts small and concrete.

  1. 01

    Software licence

    You license our control software for your own robot platform.

  2. 02

    Joint development

    We work together from concept design and simulation to a working prototype.

  3. 03

    Physical integration

    We bring our software layer onto your existing hardware and sensors.

  4. 04

    Pilot project

    We test a scoped application in a real environment, with clear criteria.

Build this robot assistant with us

We are looking for robot manufacturers, integration partners and investors to turn software into a working system.