Pedestrian–automated vehicle (AV) encounters in shared spaces often involve hesitation and ambiguity. Vehicle-mounted external human–machine interfaces (eHMIs) can help, but obscured or poorly timed communications create significant challenges. To address this, we present a mobile smart pole interaction unit (SPIU) with integrated cameras and LED displays, designed as a pedestrian-side system to deliver explicit cues (“WALK,” “STOP”). An in-the-wild evaluation of the SPIU (N = 21) using a four-factor analysis (CarBehavior, Mobility, eHMI, SPIU) showed that the SPIU improved understandability, trust, and perceived safety, and reduced workload compared with the baseline, with a combination (eHMI+SPIU) yielding the strongest results. Beyond these quantitative benefits, participants appreciated the mobility of the SPIU for its “clear” and “easy to decide” mediation. This work contributes to (1) a design and deployment framework for a mobile SPIU and (2) an in-the-wild evaluation protocol for pedestrian–AV interactions in nonsignalized spaces. Our work sparks discussions on real world evaluations involving detailed vehicle kinematics and accessible multimodality (e.g., audio), focusing on the role of personal robots as user-side eHMIs.

Field studies are irreplaceable but costly, time-consuming, and error-prone, which need careful preparation. Inspired by rapid-prototyping in manufacturing, we propose a fast, low-cost evaluation method using Vision-Language Model (VLM) personas to simulate outcomes comparable to field results. While LLMs show human-like reasoning and language capabilities, autonomous vehicle (AV)-pedestrian interaction requires spatial awareness, emotional empathy, and behavioral generation. This raises our research question: To what extent can VLM personas mimic human responses in field studies? We conducted parallel studies: 1) one real-world study with 20 participants, and 2) one video-study using 20 VLM personas, both on a street-crossing task. We compared their responses and interviewed five HCI researchers on potential applications. Results show that VLM personas mimic human response patterns (e.g., average crossing times of 5.25 s vs. 5.07 s) lack the behavioral variability and depth. They show promise for formative studies, field study preparation, and human data augmentation.

This paper introduces Multi-PrefDrive, a framework that significantly enhances LLM-based autonomous driving through multidimensional preference tuning. Aligning LLMs with human driving preferences is crucial yet challenging, as driving scenarios involve complex decisions where multiple incorrect actions can correspond to a single correct choice. Traditional binary preference tuning fails to capture this complexity. Our approach pairs each chosen action with multiple rejected alternatives, better reflecting real-world driving decisions. By implementing the Plackett-Luce preference model, we enable nuanced ranking of actions across the spectrum of possible errors. Experiments in the CARLA simulator demonstrate that our algorithm achieves an 11.0% improvement in overall score and an 83.6% reduction in infrastructure collisions, while showing perfect compliance with traffic signals in certain environments. Comparative analysis against DPO and its variants reveals that Multi-PrefDrive’s superior discrimination between chosen and rejected actions, which achieving a margin value of 25, and such ability has been directly translates to enhanced driving performance. We implement memory-efficient techniques including LoRA and 4-bit quantization to enable deployment on consumer-grade hardware and will open-source our training code and multi-rejected dataset to advance research in LLM-based autonomous driving systems. Project Page (https://liyun0607.github.io/).

The deployment of roadside LiDAR sensors plays a crucial role in the development of Cooperative Intelligent Transport Systems (C-ITS). However, the high cost of LiDAR sensors necessitates efficient placement strategies to maximize detection performance. Traditional roadside LiDAR deployment methods rely on expert insight, making them time-consuming. Automating this process, however, demands extensive computation, as it requires not only visibility evaluation but also assessing detection performance across different LiDAR placements. To address this challenge, we propose a fast surrogate metric, the Entropy-Guided Visibility Score (EGVS), based on information gain to evaluate object detection performance in roadside LiDAR configurations. EGVS leverages Traffic Probabilistic Occupancy Grids (TPOG) to prioritize critical areas and employs entropy-based calculations to quantify the information captured by LiDAR beams. This eliminates the need for direct detection performance evaluation, which typically requires extensive labeling and computational resources. By integrating EGVS into the optimization process, we significantly accelerate the search for optimal LiDAR configurations. Experimental results using the AWSIM simulator demonstrate that EGVS strongly correlates with Average Precision (AP) scores and effectively predicts object detection performance. This approach offers a computationally efficient solution for roadside LiDAR deployment, facilitating scalable smart infrastructure development.

This paper presents PrefDrive, a novel frame-work that integrates driving preferences into autonomous driving models through large language models (LLMs). While recent advances in LLMs have shown promise in autonomous driving, existing approaches often struggle to align with specific driving behaviors (e.g., maintaining safe distances, smooth acceleration patterns) and operational requirements (e.g., traffic rule compliance, route adherence). We address this challenge by developing a preference learning framework that combines multimodal perception with natural language understanding. Our approach leverages Direct Preference Optimization (DPO) to fine-tune LLMs efficiently on consumer-grade hardware, making advanced autonomous driving research more accessible to the broader research community. We introduce a comprehensive dataset of 74,040 sequences, carefully annotated with driving preferences and driving decisions, which, along with our trained model checkpoints, is made publicly available https://github.com/LiYun0607/PrefDrive/ to facilitate future research. Through extensive experiments in the CARLA simulator, we demonstrate that our preference-guided approach significantly improves driving performance across multiple metrics, including distance maintenance and trajectory smoothness. Results show up to 28.1% reduction in traffic light violations and 8.5% improvement in route completion while maintaining appropriate distances from obstacles. The framework demonstrates robust performance across different urban environments, showcasing the effectiveness of preference learning in autonomous driving applications.

As autonomous vehicles (AVs) enter pedestrian-centric environments, existing vehicle-mounted external human–machine interfaces (eHMIs) often fall short in shared spaces due to line-of-sight limitations, inconsistent signaling, and increased decision latency on pedestrians. To address these challenges, we introduce the Smart Pole Interaction Unit (SPIU), an infrastructure-based eHMI that decouples intent signaling from vehicles and provides context-aware, elevated visual cues. We evaluate SPIU using immersive VR-AWSIM simulations in four high-risk urban scenarios: four-way intersections, autonomous mixed traffic, blindspots, and nighttime crosswalks. The experiment was developed in Japan and replicated in Norway, where forty participants engaged in 32 trials each under both SPIU-present and SPIU-absent conditions. Behavioral (response time) and subjective (acceptance scale) data were collected. Results show that SPIU significantly improves pedestrian decision-making, with reductions ranging from 40% to over 80% depending on scenario and cultural context, particularly in complex or low-visibility scenarios. Cross-cultural analyses highlight SPIU’s adaptability across differing urban and social contexts. We release our open-source Smartpole-VR-AWSIM framework to support reproducibility and global advancement of infrastructure-based eHMI research through reproducible and immersive behavioral studies.

The commercialization of 5G has been initiated for a while. Furthermore, millimeter wave (mmWave) has been introduced to small cells with small coverage due to its strong linearity and non-winding characteristics. On the other hand, in connected autonomous vehicles (CAV s), where various traffic systems can cooperatively perform recognition, decision-making, and execution, communication is assumed to be always connected. Therefore, to use low latency mm Wave for high-speed moving CAV, existing beamforming cannot follow them at high speed. This paper proposes an improved beam tracking algorithm for high-speed CAVs, which can be evaluated in a more general environment using a traffic simulator. We proposed an adaptive algorithm for a general road environment by increasing the number of beam searches and search dimensions.

While text-based external human-machine interface (eHMI) is widely accepted, one limitation is the lack of capability to communicate spatial information such as a different person or location. We built a mixed-eHMI using "eye" as a target-specifier when "text" shows the clear intention to their communication partners. We conducted a pre-experimental observation to develop two testbed scenarios, followed by a video-based user study via life-size projection with a real-car prototype mounted a text display and a set of robotic eyes. The results demonstrated that our proposed "text + eye" combination may represent geospatial information by increasing the success pick-up rate.

Lane merging is one of the critical tasks for self-driving cars, and how to perform lane-merge maneuvers effectively and safely has become one of the important standards in measuring the capability of autonomous driving systems. However, due to the ambiguity in driving intentions and right-of-way issues, the lane merging process in autonomous driving remains deficient in terms of maintaining or ceding the right-of-way and attributing liability, which could result in protracted durations for merging and problems such as trajectory oscillation. Hence, we present a rule-compliance path planner (RCPP) for lane-merge scenarios, which initially employs the extended responsibility-sensitive safety (RSS) to elucidate the right-of-way, followed by the potential field-based sigmoid planner for path generation. In the simulation, we have validated the efficacy of the proposed algorithm. The algorithm demonstrated superior performance over previous approaches in aspects such as merging time (Saved 72.3%), path length (reduced 53.4%), and eliminating the trajectory oscillation.

The interactive decision-making in multi-agent autonomous racing offers insights valuable beyond the domain of self-driving cars. Mapless online path planning is particularly of practical appeal but poses a challenge for safely overtaking opponents due to the limited planning horizon. To address this, we introduce RaceMOP, a novel method for mapless online path planning designed for multi-agent racing of F1TENTH cars. Unlike classical planners that rely on predefined racing lines, RaceMOP operates without a map, utilizing only local observations to execute high-speed overtaking maneuvers. Our approach combines an artificial potential field method as a base policy with residual policy learning to enable long-horizon planning. We advance the field by introducing a novel approach for policy fusion with the residual policy directly in probability space. Extensive experiments on twelve simulated racetracks validate that RaceMOP is capable of long-horizon decision-making with robust collision avoidance during overtaking maneuvers. RaceMOP demonstrates superior handling over existing mapless planners and generalizes to unknown racetracks, affirming its potential for broader applications in robotics. Our code is available at http://github.com/raphajaner/racemop.

Pseudonyms are widely used in Cooperative Intelligent Transport Systems (C-ITS) to protect the location privacy of vehicles. However, the unlinkability nature of pseudonyms also enables Sybil attacks, where a malicious vehicle can pretend to be multiple vehicles at the same time. In this paper, we propose a novel protocol called zero-knowledge Proof of Distinct Identity (zk-PoDI,) which allows a vehicle to prove that it is not the owner of another pseudonym in the local area, without revealing its actual identity. Zk-PoDI is based on the Diophantine equation and zk-SNARK, and does not rely on any specific pseudonym design or infrastructure assistance. We show that zk-PoDI satisfies all the requirements for a practical Sybil-resistance pseudonym system, and it has low latency, adjustable difficulty, moderate computation overhead, and negligible communication cost. We also discuss the future work of implementing and evaluating zk-PoDI in a realistic city-scale simulation environment.

The rapid increase in the number of connected vehicles on roads has made vehicular ad-hoc networks (VANETs) an attractive target for malicious actors. As a result, VANETs require secure data transmission to maintain the network’s integrity. Federated learning (FL) has been proposed as a secure data-sharing method for VANETs, but it is limited in its ability to protect sensitive data. This article proposes integrating Blockchain technology into FL to provide an additional layer of security for VANETs. In particular, we propose a secure and efficient blockchain-based FL (SEBFL) approach to ensure communication efficiency and data privacy in VANETs. To this end, we use the FL model for VANETs, where computation tasks are decomposed from a base station to individual vehicles. This effectively reduces the congestion delay and communication overhead. Integrating blockchain with the FL model provides a reliable and secure data communication system between vehicles, roadside units, and a cloud server. Additionally, we use a homomorphic encryption system (HES) that effectively preserves the confidentiality and credibility of vehicles. Besides, the proposed SEBFL leverages the asynchronous FL model, minimizing the long delay while avoiding possible threats and attacks using HES. The experimental results show that the proposed SEBFL achieves 0.87% accuracy while a model inversion attack and 0.86% accuracy while a membership inference attack.

Path planning is critical for autonomous vehicles (AVs) to determine the optimal route while considering constraints and objectives. The potential field (PF) approach has become prevalent in path planning due to its simple structure and computational efficiency. However, current PF methods used in AVs focus solely on the path generation of the ego vehicle while assuming that the surrounding obstacle vehicles drive at a preset behavior without the PF-based path planner, which ignores the fact that the ego vehicle's PF could also impact the path generation of the obstacle vehicles. To tackle this problem, we propose a PF-based path planning approach where local paths are shared among ego and obstacle vehicles via vehicle-to-vehicle (V2V) communication. Then by integrating this shared local path into an objective function, a new optimization function called interactive speed optimization (ISO) is designed to allow driving safety and comfort for both ego and obstacle vehicles. The proposed method is evaluated using MATLAB/Simulink in the urgent merging scenarios by comparing it with conventional methods. The simulation results indicate that the proposed method can mitigate the impact of other AVs' PFs by slowing down in advance, effectively reducing the oscillations for both ego and obstacle AVs.

Making safe and successful lane changes (LCs) is one of the many vitally important functions of autonomous vehicles (AVs) that are needed to ensure safe driving on expressways. Recently, the simplicity and real-time performance of the potential field (PF) method have been leveraged to design decision and planning modules for AVs. However, the LC trajectory planned by the PF method is usually lengthy and takes the ego vehicle laterally parallel and close to the obstacle vehicle, which creates a dangerous situation if the obstacle vehicle suddenly steers. To mitigate this risk, we propose a time-to-collision-aware LC (TTCA-LC) strategy based on the PF and cubic polynomial in which the TTC constraint is imposed in the optimized curve fitting. The proposed approach is evaluated using MATLAB/Simulink under high-speed conditions in a comparative driving scenario. The simulation results indicate that the TTCA-LC method performs better than the conventional PF-based LC (CPF-LC) method in generating shorter, safer, and smoother trajectories. The length of the LC trajectory is shortened by over 27.1%, and the curvature is reduced by approximately 56.1% compared with the CPF-LC method.

For cooperative intelligent transport systems (C-ITS), vehicle-to-everything (V2X) communication is utilized to allow autonomous vehicles to share critical information with each other. We propose AutowareV2X, an implementation of a V2X communication module that is integrated into the autonomous driving (AD) software, Autoware. AutowareV2X provides external connectivity to the entire AD stack, enabling the end-to-end (E2E) experimentation and evaluation of connected autonomous vehicles (CAV). The Collective Perception Service was also implemented, allowing the transmission of Collective Perception Messages (CPMs). A dual-channel mechanism that enables wireless link redundancy on the critical object information shared by CPMs is also proposed. Performance evaluation in field experiments has indicated that the CPM-based perception information can be transmitted in around 30 ms, and shared object data can be used by the AD software to conduct collision avoidance maneuvers. The dual-channel delivery of CPMs transmits perception information through two different wireless technologies. The receiver-side CAV can then dynamically select the best CPM from CPMs received from both links, depending on the freshness of their information.

A high definition (HD) map, which provides prior knowledge to autonomous driving tasks, has been attracted in recent years. An HD map-based self-localization is a crucial technology for autonomous driving, but its accuracy is greatly affected by occlusions caused by dynamic obstacles in real environments. This paper focuses on clarifying the need for HD maps for stable self-localization in highly dynamic environments, especially in an urban canyon. By comparing the effects of occlusion with synthetically generated obstacles in a real environment, we show significant accuracy degradations in a general self-localization method due to obstacles in Shinjuku, Tokyo, Japan. In addition, we reveal that pole-like objects can be vital elements of an HD map to stabilize self-localization accuracy even with many obstacles by evaluating various patterns of high occlusion cases.

Map-matching based on light detection and ranging (LiDAR) is a promising method for accurate self-localization and recently has gained a wider focus due to the availability of high definition (HD) maps and price-down of LiDARs. In this method, the input scan of the LiDAR is matched to the prebuilt map to get a centimeter-level accuracy position of the vehicle. However, in some places of the map, due to the lack of features, the presence of the repetitive features, the layout of the features, and other factors, the map-matching error might exceed the required bound for autonomous driving. In our previous work, four criteria for evaluation of the features of the map was introduced and it is shown that by examining the corresponding factors for each criterion, the map-matching error can be modeled. In this work, one of the map criteria called local similarity is further investigated and in order to quantify the fulfillment of this criterion, three new factors, namelypfh_similarity,pfh_entropy, andbattacharya_similarityare introduced. In addition to this, a framework for pre-estimation of the map-matching error considering these four criteria based on random forest regression is proposed. To evaluate the accuracy of the framework, experiments were conducted for 3.6 km in Shinjuku, Tokyo. Experimental results show that using the proposed framework, in 64.1% of the cases, the localization error can be estimated with less than 2.5cm of the estimation error.

Vehicle self-Iocalization based on the matching of Light detection and ranging (LiDAR) scans to the normal distribution (ND) map become more popular in recent years due to the price down and miniaturization of the LiDARs. In such methods, the source of self-Iocalization error can be divided into input scan quality, matching algorithm and map. In this work, we focus on the map, as one ofthe high potential sources of error. By investigating the erroneous scenarios in the map and comparing their characteristics, we come up with some criteria and requirements for the map to be able to perform self-Iocalization with a needed error. In this work, we propose four factors for quantified evaluation ofthe map requirements. These factors are feature count factor, layout factor, normal entropy factor, and local similarity factor ofthe map. We evaluated these four factors in a different part ofthe map with different scenarios by comparing them with the self-Iocalization error. Experimental results show that the local similarity factor with 0.59 of correlation with the maximum error has the highest contribution to the Iocalization error. For normal entropy factor, feature count factor, layout factor, correlations are 0.42, 0.36, and 0.34 respectively. By applying these four factors, maximum Iocalization error can be modeled with RMSE and R-squared (R2) of 0.44 and 0.598 respectively. Result of this study can be applied to the dynamic determination of the abstraction ratio of the map and sensor fusion as well.

The precise map is the main provider of static environment information for the intelligent vehicles. Therefore, it is considered as a fundamental requirement for such systems. The accuracy of Mobile Mapping Systems (MMS), one of the main vehicle-based 3D laser scanning technologies, is significantly degraded due to the blockage of GPS signals in deep urban areas where tall buildings are surrounding streets. Existing solutions for the adjustment of the MMS data which require a manual measurement of the Ground Control Points (GCP) are labor-intensive and costly. In this paper, a fully-automatic framework for the calibration of the MMS is presented which corrects the 3D laser scanning data based on the road markings extracted from the aerial surveillance data. The proposed framework consists of three main steps: road marking extraction from aerial data, road marking extraction from the MMS point cloud, and the registration of the MMS road markings to the aerial reference. For the registration, a method based on the dynamic sliding window is introduced. The experimental results of the Hitotsubashi intersection in Tokyo demonstrate that the proposed method is practical for the MMS calibration in the urban area and it could achieve a pixel-level accuracy, where the Ground Sampling Distance (GSD) of the airborne image was 12cm.

Accurate vehicle self-localization is one of the crucial requirements of Autonomous vehicles. Since GPS-based localization techniques cannot achieve required accuracy in urban canyons, recently LiDAR-based (Light Detection and ranging) localization techniques gained a focus due to its accuracy. One of the challenges of LiDAR-based map matching methods is a size of the map. This paper proposes a new structure of map which is a multilayer 2D vector map and localization methods based on multi-channel LiDAR. Proposed map is extremely small in size comparing to 3D point cloud maps while preserving the localization accuracy. As this 2D map is generated by accumulating different layers of buildings, it has less uncertainty. Further, this map provides more features for map matching comparing to the conventional 2D maps and as a result, the accuracy of localization is improved. On the other hand, vector structure of the map bring more precise NDT (normal distribution transform) representation and as a result, more accurate matching. Experimental results show that proposed method outperform the conventional 2D map matching techniques in terms of accuracy.