Matej Mertik

Research profile

Research summary

Matej Mertik is a Slovenia-based researcher at Alma Mater Europaea whose OpenAlex-resolved profile spans applied machine learning and, more recently, neuromorphic computing. His topic assignments blend data mining and machine-learning applications (6 works), artificial intelligence in healthcare (5), superconducting materials and applications (5), engineering and information technology (5), and advanced neural network applications (4), pointing to an applied, cross-domain career. The current research program captured in Evolve's catalog concentrates on spiking neural networks: a 2026 arXiv preprint trains an actor-critic PPO agent to fly a UAV through constrained window openings in civil infrastructure, and a companion event-based vision study classifies pedestrian crossing intent with convolutional SNNs on synthetic DVS data. His field distribution splits evenly (0.5/0.5) between neuromorphic reinforcement learning for robotics and computer vision for autonomous driving. Available metrics include an h-index and i10-index of 7, with 7 works in the last three years and 5 distinct co-authors. Evolve's reproducibility assessment of his two reviewed papers averages 78/100, and both preprints list him as last author, consistent with a senior or supervisory role.

Research keywords

Highlights

• Spiking actor-critic PPO agent for UAV flight through building openings

  A 2026 arXiv preprint trains a spiking-neural-network actor-critic PPO controller to navigate a UAV through constrained window openings in civil infrastructure and buildings, using curriculum learning in simulation. Evolve rates reproducibility readiness high but flags 63.77% single-seed training success, no ANN baseline, and no energy data as preliminary.

• Event-based pedestrian crossing intent classification with convolutional SNNs

  In a companion 2026 study, a convolutional spiking neural network with temporal augmentation classifies pedestrian crossing intent from event-based (DVS) vision simulated with the v2e simulator, reporting 95.8% accuracy on JAAD crossing-intent data. Evolve rates reproducibility medium, cautioning that overlapping-window splits risk train-test leakage despite a well-documented pipeline.

• A coherent neuromorphic perception-and-control program

  The two reviewed papers pair event-based perception with spiking-network control, and Evolve's reproducibility mean of 78/100 across them is the strongest quantitative signal in this profile — ahead of output and impact metrics that remain unscored pending more data.

Timeline

1. 2026Last-author arXiv preprint on spiking neural network actor-critic PPO control for autonomous UAV navigation through constrained openings in civil infrastructure and buildings.
2. 2026Last-author arXiv preprint on convolutional spiking neural networks with temporal augmentation classifying pedestrian crossing intent from event-based vision.
3. 2026OpenAlex logs 5 works this year — the highest annual count since 2017's 5 — marking the peak of a re-accelerating publication cadence.

Papers in the Evolve catalog

• 2026-09-20

  Spiking Neural Network Actor-Critic Proximal Policy Optimization Control for Autonomous UAV Navigation Through Constrained Openings in Civil Infrastructure and Buildings

• 2026-09-11

  Pedestrian Crossing Intent Classification From Event-Based Vision Using Convolutional Spiking Neural Networks With Temporal Augmentation

Frequent co-authors