For investors

Barycenter Lab is a robotics lab in Berlin, founded in 2026 by three engineers. Our first robot, Kiko, is a small home robot for one person. It knows its owner, stays close, and keeps what it sees on the robot.

What we believe

Robots are moving into homes. Most of them do not know who they are for. A follow-me robot drives to the nearest person. A home camera uploads what it sees.

We think a robot that lives with you has to know you, stay with you, and keep what it sees at home. That is the whole company.

Why now

Compute. The whole stack, vision, speech and behaviour, runs today on a Jetson Orin Nano Super: 67 TOPS at 25 watts, developer kit at a 249 dollar list price. The same kit cost 499 dollars and did 40 TOPS when it launched in 2023.

The category is opening. In 2026 the first home robots go on sale, at the top end for 20,000 dollars. Kiko sits at the other end: small, one job, done well.

Privacy is a product feature. A camera robot in a home is a data protection question first, in Europe above all. Kiko answers it in hardware: nothing leaves the robot. That is easier to build and to sell from Berlin than from anywhere else.

  1. 1.NVIDIA Jetson Orin Nano Super announcement (Phoronix, 17 December 2024)
  2. 2.1X NEO preorders (Sifted, 28 October 2025)

What works today.

Every feature shows how far it really is. Nothing moves up a level before it runs on the robot.

Working

Working
  • Knows its owner (face and body, learned at setup)
  • Stays with you at a set distance, through turns and past other people
  • Answers to its name with a short sentence
  • Shows what it is doing on its face
  • Vision, speech and behaviour run on the robot
  • Depth vision on the GPU

In testing

In testing
  • Records a 60 second reel on command
  • Camera mast rises, turns a full circle, lowers
  • Keeps you in frame
  • Touch on head, left and right
  • About 70 minutes per charge
  • Other faces are not saved

Planned

Planned
  • Publish the owner recognition measurements
  • Owner app with live view, reels, battery and moods
  • Kiko v2, smaller and cheaper

Nothing leaves the robot

Everything stays on KikoStays on KikoOn-device loop01Perceivethe room02Understandwho is who03Decidewhat next04Actfollow, answer, filmOwner stays known. Data stays here.ServerNo uplink

Scenarios we test

Every recording follows a script that makes the owner hard to keep. We publish the numbers when they are real.

  • 01

    Turn around

    The owner turns away. No face to see.

    In progress
  • 02

    Jacket change

    Same person, different clothes.

    In progress
  • 03

    Occlusion

    Someone walks between Kiko and its owner.

    In progress
  • 04

    Similar clothing

    Two people in the same outfit.

    In progress
  • 05

    Leave and return

    The owner leaves the room and comes back.

    In progress
  • 06

    Low light

    Evening light, lamps only.

    In progress

What we measure

MetricValue
Identity switchesMeasuring
Time to re-find the ownerMeasuring
Time the owner is lostMeasuring
Follow-distance errorMeasuring
Frame rate on the robotMeasuring

Measuring on the robot. Numbers appear here first, before any paper.

Where we are going

  1. Now

    Kiko v1 in Berlin

    Building the first Kiko in Berlin. Owner recognition, following, voice and a face.

  2. Next

    Publish what we measured

    We publish how Kiko finds its owner and what we measured.

  3. Later

    Owner app and Kiko v2

    An owner app, then a smaller and cheaper Kiko shaped by what v1 teaches.

The team

  • Aryan Kumar Singh

    Aryan Kumar Singh

    Co-founder · AI and Robotics Lead

    Builds Kiko end to end: the body in Fusion 360 and on the 3D printer, the boards and wiring, and the perception and behaviour stack on a Jetson running ROS 2. Before Barycenter Lab he was Founding AI Engineer at EMRA AI and Product Manager and Technical Lead at Mioneer. MSc Artificial Intelligence, Berlin. Two IEEE papers.

  • Khasim Bin Saleh

    Khasim Bin Saleh

    Co-founder · AI Engineer

    Builds Kiko's owner recognition: finding people in the depth image, tracking them over time and telling the owner apart from everyone else, all on the Jetson. Before Kiko he shipped an AI tutor to more than 1,000 beta users as a machine learning intern at Coschool and built retrieval pipelines for language models. He also builds Klindok, offline documentation software for German medical practices. Co-inventor on a patent application for a connected pet feeder (2023). MSc Computer Science with a focus on Big Data and AI, SRH Berlin.

  • Saurabh Kumar

    Saurabh Kumar

    Co-founder · Robotics and Electronics Engineer

    Electrical engineer, now a robotics and AI researcher.