Learning the room
Mapping a synthetic indoor environment and checking whether localisation stays stable while the platform turns.
Berserki Robotics · Project 01
FREKI 01
The first machine from Berserki Robotics. An embodied-AI system learning to observe, navigate, and operate safely in the physical world.
In the forgeDigital model v0.1Not deployed
See the test plan PlatformTracked mobile robot
Current phaseModel and simulation
Public recordEvidence, not demo reels
From the training floor
Show the work. Label the uncertainty.
This page will collect short clips and readable results from simulation and physical tests. Until the first run exists, these are the planned baselines—not completed training.
Choosing a safe path
Planning a route around static obstacles without treating a clean simulation as proof of real-world safety.
Knowing where the floor ends
Testing drop-edge detection and the stop conditions that must work before Freki is allowed to roam unattended.
Evidence protocol
A robot is what survives contact with the room.
We separate what is known from what is assumed. No polished clip gets to erase an unstable run, an unsafe stop, or a result that has not been reproduced.
- Verified
- Vendor chassis dimensions are normalized in the digital model, and the generated ROS frame tree passes structural checks.
- Provisional
- Mass, sensor poses, centre of gravity, track friction, and drive behaviour remain estimates until the physical system can be measured.
- Failed
- Nothing is hidden here. Failed simulation runs and unsafe behaviours will be logged once testing begins.
- Next test
- Import the first model into Isaac Sim, verify motion and sensor frames, then establish a repeatable baseline navigation run.
Built in public — but not in full
We will publish useful lessons, test evidence, and selected training footage. Private environments, exact procurement, and sensitive implementation details stay off the page.