Maciej Wielgosz

Research profile

Research summary

Maciej Wielgosz, affiliated with Alma Mater Europaea in Slovenia, works at the intersection of neuromorphic computing and autonomous systems. The OpenAlex-derived record (identity: openalex; ORCID 0000-0002-4401-2957) spans two decades, from a first work in 2006 to a current high-activity phase: Evolve scores activity at the maximum 100, with 46 works in the most recent three-year window and a sharp rebound from 2 works in 2020 to 18 in 2023, 17 in 2025, and 20 in 2026. Aggregates on record include an h-index of 16 and an i10-index of 27. The profile splits evenly (0.5/0.5) between robotics/neuromorphic reinforcement learning and computer vision for autonomous driving, and Evolve's focus score of 0 reflects that breadth across five topic clusters, led by advanced neural network applications (37), embedded systems design techniques (27), and autonomous vehicle technology and safety (22). Two 2026 arXiv papers in the Evolve catalog show the current thread: a spiking-network actor-critic PPO controller flying a UAV through window openings in simulation (63.77% single-seed success; rated preliminary), and a convolutional SNN classifying pedestrian crossing intent from event-based vision (claimed 95.8% JAAD accuracy; leakage risk flagged). Reproducibility averages 78 across the two reviewed papers. Collaboration is modest — 5 distinct coauthors, collaboration score 36 — with recurring partners Matej Mertik, Francis Noah Walugembe, Tomaž Goričan, Henok Teklu, and Mustafa Sakhai.

Research keywords

Highlights

• Spiking actor-critic PPO control for UAV flight through constrained openings

  A 2026 arXiv study (Wielgosz as middle author) trains a spiking neural network actor-critic agent with proximal policy optimization and curriculum learning to fly a UAV through window openings in civil infrastructure and buildings, in simulation, reporting 63.77% single-seed training success. Evolve calls the work preliminary — no ANN baseline or energy data — but rates reproducibility readiness high.

• Event-based pedestrian crossing intent classification with convolutional SNNs

  In a second 2026 arXiv paper (middle authorship), convolutional spiking neural networks with temporal augmentation classify pedestrian crossing intent from event-based vision, reporting 95.8% accuracy on JAAD-based synthetic DVS data generated with the v2e simulator. Evolve rates reproducibility medium: the pipeline is well documented, but overlapping-window splits risk train-test leakage.

Timeline

1. 2006First recorded work appears in the OpenAlex record (1 publication), starting a publishing history that now spans two decades.
2. 2017An early peak of 13 works in the yearly counts, following 12 in 2016 — the highest output until 2023.
3. 2023Output rebounds to 18 works after the quiet 2020-2022 stretch (2, 3, 2), opening the current high-activity phase.
4. 2026Two neuromorphic arXiv papers enter the Evolve catalog — SNN actor-critic PPO UAV control and event-based pedestrian intent classification — as the yearly count reaches 20 works.

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