What it does
Lithium batteries and sharps hidden in household waste cause a UK facility fire every day and injure the people sorting it. TBex is a pin-together frame of camera-and-sensor modules that clips onto any conveyor to spot hazards and remove them in time.
Your inspiration
I went into engineering to make the planet more sustainable, convinced the most efficient lever was intelligent robots doing the dirty, dangerous work done by hand today. Then I learned lithium-ion batteries hidden in household waste cause at least one facility fire every day in the UK alone, destroying equipment, poisoning air, endangering the people who sort our recycling. After which i immediately thought: This is definitely solvable with intelligent robotics! I decided to spend a substantial amount of time on it. TBest(named from my Teda heritage), is the result: a system to catch what hurts people and burns facilities, before it does.
How it works
An adjustable frame straddles the conveyor: telescoping legs, width members and a tilting rail set, locked by quick-release pins on a 25 mm hole grid, so it fits any belt's height, width and incline without tools. Sensing modules hook onto the rails. Each has a downward camera with LED lighting behind sealed windows, six swappable sensor heads (near-infrared for plastics, thermal for hot batteries) whose fields overlap so nothing slips between them, and a small computer that outlines every object and stamps its exact time. A clamp-on wheel reads belt speed, because every part indexes on the same pin grid, all positions are known by design, so no calibration is needed. A central computer matches each outline to its readings by time, identifies hazards, and fires the chosen actuator in time( air jet, robotic arm, or both). On a singulated belt (the normal feed) coverage is total. Each module's is designed to have configurable sensors or camera depending on needs.
Design process
It began with ZapSortBot (2024-25): a working prototype proving the full see-decide-intercept loop ( conveyor, camera, robot arm) published at IEEE IECON 2025 and open-sourced. It also exposed what would fail in a real facility: fixed system, one sensor, lab calibration. TBest is the redesign around those failures. After filing a UK patent (April 2026) I developed it through 11+ CAD designs iterating on the entire system after finding additional challenges and needs the current wouldn't satisfy. Every review challenged a decision and changed the hardware: the camera's small aperture vignetted the lens, so i switched to a glazed full-view window. A single centred sensor left blind strips, so it became a six-head gantry with overlapping fields. A pivot shaft blocked module seating, so it became stub trunnions with indexed lock discs. I simulated the system end-to-end from the CAD geometry: total coverage on a singulated belt; camera and sensor readings associated to 2.1 mm; interception above 95% across belt speeds. Crucially, it was accepted to ICURe Discover where i did over 90+ interviews and found strong interest at a £100k system price and a recommendation to commercialise. Three configurations were then rebuilt over real conveyor geometry: flat, inclined and twin-belt.
How it is different
Industrial optical sorters are monolithic: one belt width, one sensing method, 500k+ cost, specialist calibration, and if the sensor fails the line stops. TBest inverts every assumption, and its defining property is that it is configured, not rebuilt. For any conveyor you choose four things independently: the module payload (which cameras and sensors), the number of modules (coverage scales by count, not machine size), the compute power (sized to the fleet), and the interception method (air jet, arm, or both). One architecture serves them all. It retrofits to any belt's height, width and incline in an afternoon; a failed module swaps in minutes while neighbours keep watching; payloads change with a cheap laser-cut plate. And because every module, sensor and rail indexes on one pin grid, all positions are known by construction - the calibration that makes industrial vision expensive disappears. Protection priced for ordinary facilities, not just giants.
Future plans
Having completed ICURe Discover program with 90+ interviews showing strong demand at a £100k price and a recommendation to commercialise, the next step is to secure a paid line trial with a materials-recovery partner potentially using a module configured with an X-Ray camera, catching hot lithium cells before the shredder. The vision for TBex is to be a certified retrofit system with a chassis, sensor payloads, central unit and interception options therefore letting an ordinary facility protect a line, and its workers, for the cost of a single industrial sensor.
Awards
ZapSortBot as my final year project during my undergraduate received the highest possible grade, was peer-reviewed and published at IEEE IECON 2025 and released open-source under an MIT licence. I was selected for and completed the ICURe Discover program and was successfully recommended to progress to ICURe Explore.
Connect