2962-780X

No URL ">
  •  
  •  
 

Abstract

Static pelican crossing systems often fail to accommodate the stochastic arrival patterns prevalent in high-density transit-oriented development (TOD) zones, creating operational inefficiencies between pedestrian clearance and vehicular flow. At the Cikini Station transit hub in Jakarta, existing fixed signal cycles contribute to pedestrian delays and peak-hour congestion. This study evaluates an adaptive signal control model for mid-block pelican crossings using a Deep Q-Network (DQN) algorithm. To calibrate the simulation, empirical data were extracted from field CCTV footage using the YOLOv8 algorithm, accurately capturing 15-minute peak flow fluctuations and commuter surges. The system is formulated as a Markov Decision Process (MDP) and trained through microscopic simulation in SUMO via TraCI, utilizing real-time queue states to dynamically adjust phase durations. Results indicate that the DQN agent reduced average pedestrian waiting times by 64% (from 28.0 to 10.0 seconds) compared to the static configuration. Furthermore, by enforcing a 3-second all-red clearance and a 6-second evacuation interval, the model eliminated pedestrian risk exposure (0%). Consequently, peak-hour vehicular delay underwent a calculated increase to 33.95 seconds to prioritize safety, yet the intersection maintained a stable Level of Service (LOS). Ultimately, this research provides a data-driven framework for applying reinforcement learning to balance multi-modal mobility demands in high-variance urban traffic environments

Bahasa Abstract

Sistem pelican crossing statis sering gagal mengakomodasi pola kedatangan acak yang biasa terjadi pada zona pengembangan berorientasi transit (transit-oriented development/TOD) dengan kepadatan tinggi. Ketidakfleksibelan ini menciptakan ketidakseimbangan yang tidak proporsional antara keselamatan pejalan kaki dan arus kendaraan, sebagaimana dicontohkan oleh area berorientasi transit Stasiun Cikini, Jakarta, dimana siklus sinyal tetap yang ada memperburuk kemacetan dan risiko konflik. Studi ini mengusulkan optimasi kontrol sinyal adaptif untuk pelican crossing mid-block menggunakan pendekatan reinforcement learning Deep Q-Network (DQN). Sistem ini dimodelkan sebagai Markov Decision Process (MDP), memanfaatkan kondisi antrean pejalan kaki dan kendaraan serta waktu eksisting untuk menentukan durasi fase optimal secara dinamis. Model dilatih melalui simulasi mikroskopis di SUMO, diintegrasikan melalui TraCI, dan dikalibrasi dengan data empiris yang diekstraksi dari rekaman CCTV lapangan. Hasil menunjukkan bahwa agen DQN secara signifikan mengungguli konfigurasi statis dari kondisi eksisting. Dengan memperluas fase hijau pejalan kaki secara adaptif hingga 28 detik dan memberlakukan interval clearance all-red 3 detik dan 6 detik yang tepat, model berhasil mengurangi tingkat paparan risiko menjadi 0%. Secara kuantitatif, pendekatan RL yang diusulkan mengurangi waktu tunggu pejalan kaki rata-rata sebesar 64% (dari 28 detik menjadi 10 detik). Meskipun tundaan kendaraan meningkat menjadi 33,95 detik selama jam sibuk, hal ini tetap berada dalam Level of Service (LOS) yang stabil. Penelitian ini menunjukkan bagaimana infrastruktur berbasis AI dapat secara efektif mengubah simpul perkotaan menjadi lingkungan transit yang cerdas, aman, dan inklusif.

References

Alhajyaseen, W. and Asano, M. (2016). The Assessment of Pedestrian-Vehicle Conflicts at Crosswalks Considering Sudden Pedestrian Speed Change Events. https://doi.org/10.5339/qfarc.2016.ictpp2369

Armellini, M. (2022). Simulation of Demand Responsive Transport using a dynamic scheduling tool with SUMO. sumo Conference Proceedings, 2, 115-123. https://doi.org/10.52825/scp.v2i.100

Bouktif, S., Cheniki, A., & Ouni, A. (2021). Traffic signal control using hybrid action space deep reinforcement learning. Sensors, 21(7), 2302. https://doi.org/10.3390/s21072302

 Cao, K., Wang, L., Zhang, S., Duan, L., Jiang, G., Сфарра, С., … & Jung, H. (2024). Optimization control of adaptive traffic signal with deep reinforcement learning. Electronics, 13(1), 198. https://doi.org/10.3390/electronics13010198

 Cao, Y., et al. (2025). Intelligent connected adaptive signal control considering pedestrians based on the EXP-DDQN algorithm. PLOS One. https://doi.org/10.1371/journal.pone.0314501

Foltýnová, H. P., et al. (2024). Reflection of the sustainable urban mobility paradigm shift in teaching at three European universities. Envigogika, 18(1). https://doi.org/10.14712/18023061.658

 François-Lavet, V., Henderson, P., Islam, R., Bellemare, M., & Pineau, J. (2018). An introduction to deep reinforcement learning. Foundations and Trends® in Machine Learning, 11(3–4), 219–354. https://doi.org/10.1561/2200000071

Download

Share

COinS