Research 01Autonomous mobility
RoboZET is building an AI system that can drive a tram, starting on Zagreb's network. The first version already reads the street ahead and decides how the tram should respond.
Jump to: How it works · Phase 1 · Scout 3D · Roadmap · Contact
01 · Why trams
A tram follows fixed rails, so it never steers. The only decisions are when to accelerate, coast, brake or stop.
That turns a hard open-world problem into a narrower one: understand the scene ahead and control speed safely. Operators also need to modernise. ZET reported that only 57.47% of its trams were technically available in 2023, and it is now replacing its oldest Tatra trams.
02 · Pipeline
Driving is split into four stages, so each one can be tested and replaced on its own.
Cameras at the front capture the track and street ahead.
A local vision system finds people, cars, bikes and lights, and works out what is on or heading toward the rails.
The scene goes to Jev by TypeSafe AI, which returns a driving action with a confidence for each option.
A controller turns the action into motor and brake commands. A hardware emergency stop always overrides the AI.
03 · Phase 1
Put the AI on real tram hardware, then turn that hardware into a scout robot that works the network every day. Estimated at about 12 months.
The first version runs on recorded cab footage of ZET line 2, at about five frames per second.
The platform is a single motor bogie from a retired Tatra T4 tram, the type ZET runs as TMK 401 and is now replacing.
First runs take place on a closed test track.
Once the bogie drives reliably under AI control, it becomes a self-propelled scout. Its first job is to record a complete line to train the driving AI. After that it keeps working every day, giving the operator live, updated scans of every line.
Runs on live lines would start outside service hours, with the operator's approval.
From the first crude test rig to the final robot. All three stages share the Tatra T3 metre-gauge bogie; the parts are concept volumes, not engineered parts.
Drag to orbit · click, then scroll to zoom
The 3D model needs WebGL, which this browser has turned off.
04 · Roadmap
Three phases to a driverless line, then the network.
05 · Full ride
Ten minutes on line 2, with every detection and decision on screen.
06 · Funding
Phase 1 can be funded from one source or a combination.
EU and Croatian innovation and smart-mobility programmes for applied AI and transport.
Co-development with the operator, the city, research institutions or rail industry, through funding, hardware, track access or expertise.
Equity from investors in autonomous mobility and applied AI, backing a path from a working proof of concept to a product for tram operators across Europe.
07 · Talk to us
RoboZET is developed by Astro Synapse, a Croatian company building applied AI systems.
Decision model: Jev by TypeSafe AI. Demo footage used for research purposes.
© Astro Synapse Limited. Dublin, Ireland.