Adarsh Nanjaiya Latha wins first prize at TinyML hackathon

Last week, PS PhD student Adarsh Nanjaiya Latha participated in the ACM Europe & 4TU.NIRICT Summer School “Systems Meet AI” at TU Delft. The five-day summer school focused on the intersection of artificial intelligence and computing systems, covering topics such as edge intelligence, embedded AI and Tiny Machine Learning.

The programme concluded with a two-day hackathon on TinyML for microcontrollers. Participants worked in teams to develop a Visible Light Positioning system that could run entirely on a Raspberry Pi Pico.

The system used measurements of the brightness of overhead LEDs to estimate its position in a room at centimetre-level accuracy. All processing had to take place directly on the microcontroller, without access to cloud computing or a GPU.

The challenge was based on the DenseVLC dataset and included several scenarios, such as clean positioning, robustness to raw and noisy data, limited training samples and changes caused by LED ageing. The teams were evaluated using hidden test data, with scores based on positioning accuracy, model size, on-device inference latency and robustness.

Adarsh and his teammate, Rishi Bala Sivasubramanian, competed as Team Code Detectives and won first prize among all participating teams. Their solution demonstrated that accurate positioning could be achieved while maintaining a compact model and fast inference directly on the Raspberry Pi Pico.

Congratulations to Adarsh and Rishi on this outstanding achievement!