Accelerating Sophisticated Automated Driving System Development and Evaluation through Proactive Multi-Agent Traffic Simulation

Sou Kitajima･Shun Endo･Nobuyuki Uchida･Kunio Yamazaki (JARI)･Naoki Suganuma (Kanazawa University)･Tadashi Okuno (OS Planning)･Jun Tajima (Misaki Design)

This paper proposes a systematic evaluation methodology to accelerate the development of sophisticated Automated Driving Systems (ADS) by proactively utilizing multi-agent traffic simulation. Advanced ADSs are expected to continuously improve their collision avoidance capabilities and tactical driving performance within the operational design domain through frequent software updates. The proposed virtual testing platform enables comprehensive and iterative evaluation of integrated ADS performance under diverse and interactive traffic conditions. It supports developers in achieving reliable ADS behavior prior to real-world deployment by facilitating quantitative assessment of the system's ability to anticipate and mitigate potential safety risks and disruptions to surrounding traffic flow.

Analysis of Highway Merging Driving Behavior by Parameter Estimation of Discrete Choice based Driver Model using data from Networked Multi-Participant Driving Simulator

Tohru Yoshioka (Mazda)･Hironori Suzuki (Toyo University)･Jun Tajima (Misaki Design)

We model merging and mainline vehicle behavior on highways to support merging assistance. Using a driving simulator where ten drivers share the same virtual traffic environment, we examine interactions among vehicles and agent models under mixed traffic conditions. From merging actions involving vehicle‑to‑vehicle interactions, each driver's behavior is described within a unified modeling framework, enabling analysis of differences across vehicles.

Analysis of Individual Lane-Change Behavior and Interaction among Vehicles Caused by Lane Closure on Highway Using 10-Networked Driving Simulators

Hironori Suzuki (Toyo University)･Ryuuya Seki (Gradute School of Toyo University)･Seiya Fujii･Hiroshi Unesaki･Shuichi Kondo･Toru Yoshioka (Mazda)

This paper deals with observation and analysis of lane-change behavior of individual drivers caused by lane closure on the highway, and simultaneously, investigates the effect of resulting interactions among vehicles on traffic flow. For this purpose, we conduct simulator experiments using networked driving simulator which can offer same virtual road traffic environment to maximum of 10 drivers.

Unified Path Planning Algorithm for Traffic Participants in Multi-Agent-based Simulation (Part2)

Ryuya Seki (Graduate School of Toyo University)･Hironori Suzuki (Toyo University)･Jun Tajima (Misaki Design)

We are exploring possibility of discrete-choice-based path planning algorithm that can be uniformly applied to vehicles, motorcycles, pedestrians, bicycles, and other traffic participants appear in the multi-agent traffic simulation. As part 2 of the research report, we present the simulation result that indicates multiple agents can move narrow path interacting with each other in this paper. We also discuss how to integrate the required behavior to obey traffic rules such as stop at stopping location into our model.

Comparison of Pedestrian Behaviors Observed in Real-World Surveys and MR Pedestrian Simulator Experiments

Hironobu Hasegawa･Taiki Ago (Kagawa University)･Makoto Kasai (National Institute of Technology, Akita College)

Researchers typically understand pedestrian behavior in road environments using two main approaches: observation and experimentation. Observation enables the capture of actual behaviors but makes it difficult to obtain sufficient samples of rare events. Experimental approaches, in contrast, allow researchers to design specific scenarios, and mixed-reality (MR) simulators in particular enable the collection of pedestrian behavior under controlled conditions with embodied interaction. This study compares data obtained through these two approaches and discusses their implications for the development of pedestrian behavioral models.