Pedestrian Interaction Research Articles

SISGAN: A Generative Adversarial Network Pedestrian Trajectory Prediction Model Combining Interaction Information and Scene Information

Accurate pedestrian trajectory prediction is crucial in many fields. This requires the full use and learning of pedestrians’ social interactions, movements, and environmental information. In view of the current research on pedestrian trajectory prediction, wherein most of the pedestrian interaction information is explored from the level of overall interaction, this paper proposes the SISGAN model, which designs a social interaction module from the perspective of the target pedestrian, and takes four kinds of interaction information as the influencing factors of pedestrian interaction, so as to describe the influence mechanism of pedestrian–pedestrian interaction. In addition, in terms of environmental information, the index density of pedestrian historical trajectory in space is taken into account in the extraction of environmental information, which increases the potential correlation between environmental information and pedestrians. Finally, we integrate social interaction information and environmental information and make the final trajectory prediction based on GAN. Experiments on ETH and UCY datasets demonstrate the effectiveness of the SISGAN model proposed in this paper.

Construction of virtual test scenarios for intelligent car-pedestrian interaction

Aiming at the needs of intelligent car and pedestrian interaction testing under urban conditions, this paper proposes a scenario generation method that comprehensively considers the frequency of scenarios in the real world and the degree of challenge to human-vehicle interaction performance. First, according to the key features of intelligent car and pedestrian interaction, the original scene data of pedestrians crossing the road is extracted from the natural driving dataset; then, according to the needs of accelerated testing, a key scenario extraction method based on importance sampling theory is designed to extract and construct important scenarios for intelligent car-pedestrian interaction testing from the original scenes; finally, by comparing the data distribution of important scenarios with original scenarios, it is shown that this method can effectively screen out scenarios that may challenge safety performance during driving, thereby realizing accelerated testing, while taking into account the statistical characteristics of the test scenarios.

STAFGCN:A spatial-temporal attention-based fusion graph convolution network for pedestrian trajectory prediction

Pedestrian trajectory prediction provides crucial data support for the development of smart cities. Existing pedestrian trajectory prediction methods often overlook the different types of pedestrian interactions and the micro-level spatial-temporal relationships when handling the interaction information in spatial dimension and temporal dimension. The model employs a spatial-temporal attention-based fusion graph convolutional framework to predict future pedestrian trajectories. For the different types of local and global relationships between pedestrians, it first employs spatial-temporal attention mechanisms to capture dependencies in pedestrian sequence data, obtaining the social interactions of pedestrians in spatial contexts and the movement trends of pedestrians over time. Subsequently, a fusion graph convolutional module merges the temporal weight matrix and the spatial weight matrix into a spatial-temporal fusion feature map. Finally, a decoder section utilizes TimeStacked Convolutional Neural Networks to predict future trajectories. The final validation on the ETH and UCY datasets yielded experimental results with an Average Displacement Error(ADE) of 0.34 and an Final Displacement Error(FDE) of 0.55. The visualization results further demonstrated the rationality of the model.

The Impact of Pedestrian Interactions at Intersections on the Three Levels of Drivers’ Situation Awareness

Evaluating drivers’ situation awareness (SA) is important in the implementation of alert prioritization. This study investigates the relationship between driving performance measures (speed, acceleration and brake usage, steering wheel and lane deviation), pedestrian interaction (location, direction and motion), and driver SA. To achieve this, a controlled study was conducted with 56 participants using a Balanced Incomplete Block Design, where each participant drove 18 out of 48 possible intersections in a driving simulator environment. The Situational Awareness Global Assessment Technique (SAGAT) method was used to assess drivers’ SA. Mixed effects logit models were developed to examine the different SA Levels (perception, comprehension, projection). The driving performance measures were aggregated across three time windows (1, 3, and 5 s). The findings show significant contributions from both driving performance measures and pedestrian interactions in predicting driver SA. More specifically, a one-second time window was useful for predicting pedestrian direction and a three-second time window was best for predicting pedestrian location and intention to cross. The results indicate the importance of considering different time windows for predicting various levels of driver SA responses. These findings offer insights into factors to be considered in driver SA predictive models.

Using distributed simulations to investigate driver-pedestrian interactions and kinematic cues: Implications for automated vehicle behaviour and communication

As we move towards a future with Automated Vehicles (AVs) incorporated in the current traffic system, it is crucial to understand driver-pedestrian interaction, in order to enhance AV design and optimization. Previous research in this area, which has primarily used naturalistic observations or single-actor virtual reality simulations, has been limited by its inability to draw causal conclusions, also due to a lack of real human–human interactions. Our study addresses these limitations by employing a high-fidelity distributed simulation setup that links drivers in a motion-based simulator with pedestrians in a CAVE-based environment. This method allows for the examination of real-time and reciprocal interactions across a range of road-crossing scenarios. Using thirty-two pairs of drivers and pedestrians, we investigated how different factors, such as the presence of zebra crossings and varying time gaps of the approaching vehicle, influence driver behaviour and pedestrian crossing decisions. The effect of drivers’ control of the vehicle during such crossings (e.g., braking behaviour and lateral deviation) on pedestrians’ crossing decisions were also analysed. We found that the distribution of drivers’ average deceleration values were bimodal, where drivers either markedly yielded to pedestrians, or continued in their path, with very few instances of intermediate behaviour. We also found that pedestrian decisions were seemingly influenced by the different braking strategies adopted by the driver, with pedestrians crossing before the vehicles in response to soft and early, or late and hard braking, while late and soft braking often resulted in the vehicle passing first. We also observed a slight lateral movement of the vehicle away from pedestrians when drivers were not yielding, but more of a lateral deviation towards them when yielding. This may be because drivers subconsciously transfer their walking interaction habits to their driving behaviour, to avoid a collision with pedestrians. Finally, our results showed a stronger influence of these kinematic cues on pedestrian crossing decisions, when compared to zebra crossings. As well as highlighting the value of a novel approach for investigating vehicle–pedestrian interactions, this study illustrates how vehicle cues can assist pedestrian decisions, adding new knowledge in the development of human-like behaviour for future AVs.

Autonomous Vehicle–Pedestrian Interaction Modeling Platform: A Case Study in Four Major Cities

Autonomous Vehicle–Pedestrian Interaction Modeling Platform: A Case Study in Four Major Cities

Investigating LiDAR Sensor Accuracy for V2V and V2P Conflict Detection at Signalized Intersections

This paper examined the accuracy of six installed LiDAR sensors at six different signalized intersections in Trois-Rivières City, Quebec, Canada. At each intersection, the crucial leading and following movements that cause vehicle–vehicle (V2V) and vehicle–pedestrian (V2P) conflicts were identified, and the LiDAR results were compared to crash reports recorded by police, insurance companies, and other reliable resources. Furthermore, the intersection crash rates were calculated based on the daily entering vehicle traffic and the frequency of crashes at each intersection. Convolutional Neural Networks (CNNs) were utilized over 970,000 V2V and V2P conflicts based on the post encroachment time (PET) and time-to-collision (TTC) safety assessment measures. Bayesian models were used to assess the relationships between different intersection characteristics and the occurrence of conflicts, providing insights into the factors influencing V2V and V2P conflict occurrences. Additionally, a developed image-processing algorithm was utilized to examine the conflicts’ trajectories. The intersections’ crash rates indicated that safety considerations should be implemented at intersections #3, #6, #4, #1, #5, and #2, respectively. Additionally, intersections #6, #4, and #3 were the intersections with the highest rates of vehicle–pedestrian conflicts. Analysis revealed the intricate nature of vehicle and pedestrian interactions, demonstrating the potential of LiDAR sensors in discerning conflict-prone areas at intersections.

A novel model for real-time risk evaluation of vehicle–pedestrian interactions at intersections

Safety decisions for vehicles at an intersection rely on real-time, objective and continuous assessment of risks in vehicle–pedestrian interactions. Existing surrogate safety models, constrained by ideal assumptions of constant current speed and reliant on interaction points, often misjudge risks, and show inefficiency, inaccuracy and discontinuity. This work proposes a novel model for evaluation of those risks in vehicle–pedestrian interactions at intersections, which abstracts the pedestrian distribution density around a vehicle into a generalized model of driver-pedestrian interaction preferences. The introduction of two conceptions: ’driving risk index’ and ’driving risk gradient,’ facilitates the delineation of driving spaces for identifying safety–critical events. By means of the trajectory data from three intersections, model parameters are calibrated and a multidimensional vehicle–pedestrian interaction risk (VPIR) model is proposed to adapt the complex and dynamic characteristics of vehicle–pedestrian interactions at intersections. Commonly used surrogate safety models, such as Time to Collision (TTC), are selected as benchmark models. Results show that the proposed model overcomes the limitations of the existing interaction-point-based models, and offers a ideal assessment of driving risks at intersections. Finally, the model is illustrated with a case study that assesses the risks in vehicle–pedestrian interactions in varied scenarios and the case study indicates that the VPIR model works well in evaluating vehicle–pedestrian interaction risks. This work can facilitate humanoid learning in the autonomous driving domain, and achieve an ideal evaluation of vehicle–pedestrian interaction risks for safe and efficient vehicle navigation through an intersection.

The Impact of Pedestrian Lane Formation by Obstacles on Fire Evacuation Efficiency in the Presence of Unfair Competition

After a fire breaks out, pedestrians simultaneously move towards the exit and quickly form a crowded area near the exit. With the intensification of pedestrians’ tendencies towards unfair competition, there is an increase in pushing and collisions within the crowd. The possibility of stampedes within the crowd also gradually increases. Analyzing the causes and psychological tendencies behind pedestrian pushing and collisions has a positive effect on reducing crowd instability and improving evacuation efficiency. This research proposes a modified social force model considering the unfair competition tendency of pedestrians. The model considers factors such as the gap between pedestrians’ actual and maximum achievable speed, effective radius, and their distance from the exit. In order to overcome the shortage of “deadlock” in the classical social force model in a high-density environment, this research introduces the feature of variable pedestrian effective radius. The effective radius of pedestrians dynamically changes according to the density of the surrounding crowd and queuing time. Through validation, the evacuation efficiency of this model aligns well with the actual situation and effectively reflects pedestrians’ pushing and squeezing behaviors in high-density environments. This research also analyzes how to strategically arrange obstacles to mitigate the exacerbating effect of unfair pedestrian competition on exit congestion. Five experiments were conducted to analyze how the relative position of obstacles and exits, the number of evacuation paths, and the size of the obstacle-free area before the exit affect evacuation efficiency in the presence of unfair pedestrian competition. The results show that evacuation efficiency can be improved when obstacles play a role in guiding or reducing the interaction of pedestrians in different queues. However, when obstacles hinder pedestrians, the evacuation efficiency is reduced to a certain extent.

Interactions and Behaviors of Pedestrians with Autonomous Vehicles: A Synthesis

Integrating autonomous vehicles (AVs) into public roads presents profound implications for pedestrian safety and the broader acceptance of this emerging technology. This work examines the complex interactions between AVs and pedestrians, a dynamic influenced by the variability of pedestrian behaviors and the absence of traditional communication mechanisms, such as eye contact and gestures, commonly relied upon in human-driven scenarios. Given the nascent stage of AV deployment, this research addresses the challenges of evaluating AV−pedestrian interactions amid safety concerns and technological limitations. We review and synthesize global research on pedestrian behavior in the context of AV technology to track changes in pedestrians’ acceptance over time and identify the factors driving these shifts. Additionally, this review incorporates insights from transportation authorities to highlight potential safety issues and the need for innovative communication strategies that ensure safe interactions between pedestrians and AVs. By analyzing these factors, the research aims to contribute to the development of guidelines and communication protocols that enhance pedestrian safety and facilitate the integration of AVs into urban environments.

Model inductive bias enhanced deep reinforcement learning for robot navigation in crowded environments

Navigating mobile robots in crowded environments poses a significant challenge and is essential for the coexistence of robots and humans in future intelligent societies. As a pragmatic data-driven approach, deep reinforcement learning (DRL) holds promise for addressing this challenge. However, current DRL-based navigation methods have possible improvements in understanding agent interactions, feedback mechanism design, and decision foresight in dynamic environments. This paper introduces the model inductive bias enhanced deep reinforcement learning (MIBE-DRL) method, drawing inspiration from a fusion of data-driven and model-driven techniques. MIBE-DRL extensively incorporates model inductive bias into the deep reinforcement learning framework, enhancing the efficiency and safety of robot navigation. The proposed approach entails a multi-interaction network featuring three modules designed to comprehensively understand potential agent interactions in dynamic environments. The pedestrian interaction module can model interactions among humans, while the temporal and spatial interaction modules consider agent interactions in both temporal and spatial dimensions. Additionally, the paper constructs a reward system that fully accounts for the robot’s direction and position factors. This system's directional and positional reward functions are built based on artificial potential fields (APF) and navigation rules, respectively, which can provide reasoned evaluations for the robot's motion direction and position during training, enabling it to receive comprehensive feedback. Furthermore, the incorporation of Monte-Carlo tree search (MCTS) facilitates the development of a foresighted action strategy, enabling robots to execute actions with long-term planning considerations. Experimental results demonstrate that integrating model inductive bias significantly enhances the navigation performance of MIBE-DRL. Compared to state-of-the-art methods, MIBE-DRL achieves the highest success rate in crowded environments and demonstrates advantages in navigation time and maintaining a safe social distance from humans.

InfoSTGCAN: An Information-Maximizing Spatial-Temporal Graph Convolutional Attention Network for Heterogeneous Human Trajectory Prediction

Predicting the future trajectories of multiple interacting pedestrians within a scene has increasingly gained importance in various fields, e.g., autonomous driving, human–robot interaction, and so on. The complexity of this problem is heightened due to the social dynamics among different pedestrians and their heterogeneous implicit preferences. In this paper, we present Information Maximizing Spatial-Temporal Graph Convolutional Attention Network (InfoSTGCAN), which takes into account both pedestrian interactions and heterogeneous behavior choice modeling. To effectively capture the complex interactions among pedestrians, we integrate spatial-temporal graph convolution and spatial-temporal graph attention. For grasping the heterogeneity in pedestrians’ behavior choices, our model goes a step further by learning to predict an individual-level latent code for each pedestrian. Each latent code represents a distinct pattern of movement choice. Finally, based on the observed historical trajectory and the learned latent code, the proposed method is trained to cover the ground-truth future trajectory of this pedestrian with a bi-variate Gaussian distribution. We evaluate the proposed method through a comprehensive list of experiments and demonstrate that our method outperforms all baseline methods on the commonly used metrics, Average Displacement Error and Final Displacement Error. Notably, visualizations of the generated trajectories reveal our method’s capacity to handle different scenarios.

Vibration serviceability of footbridges in crowded conditions: crowd dynamics simulations vs guidelines’ predictions

Abstract Vibration serviceability assessment in crowded conditions requires a reliable human-induced loading model taking into account pedestrian interaction. Current guidelines provide simplified procedures to determine the maximum dynamic response based on very simplified loading models. The main objective of this paper is to assess the reliability of current guidelines for vibration serviceability assessment of footbridges. With this aim, an extensive campaign of numerical simulations of bidirectional pedestrian traffic is carried out through an agent-based model, considering variable pedestrian densities and deck widths. The results of numerical simulations are first compared against the experiments available in the literature in terms of fundamental diagram of the mean walking speed as a function of pedestrian density. Then, starting from numerical simulations, the dynamic response of a class of footbridges is estimated numerically and compared with guidelines’ predictions in order to assess their reliability.

Coupling intention and actions of vehicle–pedestrian interaction: A virtual reality experiment study

The interactions between vehicles and pedestrians are complex due to their interdependence and coupling. Understanding these interactions is crucial for the development of autonomous vehicles, as it enables accurate prediction of pedestrian crossing intentions, more reasonable decision-making, and human-like motion planning at unsignalized intersections. Previous studies have devoted considerable effort to analyzing vehicle and pedestrian behavior and developing models to forecast pedestrian crossing intentions. However, these studies have two limitations. First, they mainly focus on investigating variables that explain pedestrian crossing behavior rather than predicting pedestrian crossing intentions. Moreover, some factors such as age, sensation seeking and social value orientation, used to establish decision-making models in these studies are not easily accessible in real-world scenarios. In this paper, we explored the critical factors influencing the decision-making processes of human drivers and pedestrians respectively by using virtual reality technology. To do this, we considered available kinematic variables and analyzed the internal relationship between motion parameters and pedestrian behavior. The analysis results indicate that longitudinal distance and vehicle acceleration are the most influential factors in pedestrian decision-making, while pedestrian speed and longitudinal distance also play a crucial role in determining whether the vehicle yields or not. Furthermore, a mathematical relationship between a pedestrian’s intention and kinematic variables is established for the first time, which can help dynamically assess when pedestrians desire to cross. Finally, the results obtained in driver-yielding behavior analysis provide valuable insights for autonomous vehicle decision-making and motion planning.

The Development of Modeling Shared Spaces to Support Sustainable Transport Systems: Introduction to the Integrated Pedestrian–Vehicle Model (IPVM)

The significance of developing shared road infrastructure in cities throughout the world is growing. Driven by the need to improve traffic management in ways that enhance multiple sustainability outcomes, developing the tools needed to test shared space proposals is becoming more sought after by responsible agencies. This paper reviews approaches to simulation modeling focused on representing and assessing shared spaces, culminating in a new approach presented here called the Integrated Pedestrian–Vehicle Model (IPVM)—a novel framework that combines social force models, car-following models and other algorithms from the robotics domain to better describe both mobility and activity within a shared space. The IPVM recognizes that while shared spaces are inherently multimodal, past efforts have tended to use pedestrian models as a starting point. Most consider the interaction of pedestrians with other pedestrians and static road infrastructure. Shared space models are generally microscopic models that integrate a social force model with a variety of car-following models to describe the interaction between vehicles and pedestrians. However, there is little research and few practical methodologies that address the long-range conflict avoidance between vehicles and pedestrians. This aspect is crucial for accurately representing the desire lines and pathways of pedestrians and active transport users in complex environments like shared spaces. The IPVM describes and visualizes shared road infrastructure with an absence of separating infrastructure between users and outputs. It generates metrics that can be used in conjunction with the latest evaluation approaches to gauge the sustainability credentials of shared space road proposals. Enhanced modeling of shared space solutions can lead to more effective implementation, which can potentially reduce the presence of cars, increase public and active transport use and lead to a more sustainable transport system.

A modified social force model for pedestrian-bicycle mixed flows and its application on evaluating the conflict risk in shared roads

As sustainable modes of transport, walking and cycling are the major choices for short trips. The mixed flow of pedestrians and bicycles is often observed on the roads, but its dynamics and conflict risks have not been well investigated. As commonly observed, pedestrians walk along the road boundaries in the pedestrian-bicycle shared roads. When two or more bicycles share the road, the rear bicycle may choose to either follow or overtake the preceding bicycle under specific criteria. To better reproduce this phenomenon, this study proposed a modified social model. Specifically, self-driving force, force from boundaries, force from other pedestrians and force from bicycles were considered in simulating pedestrian movement. Additionally, the modeling of bicycle movement explicitly takes into account the behavior force for following or overtaking the preceding bicycle. YOLO v5 object detection and DeepSORT multi-object tracking algorithms were applied to extract pedestrian and bicycle trajectories captured by the camera on an unmanned aerial vehicle (UAV). Then the model was calibrated using a genetic algorithm to minimize the discrepancy between the simulated and observed trajectories. Under both unidirectional and bidirectional flow scenarios, the proposed model demonstrates good accuracy in reproducing individual movements and lane-formation phenomena for bidirectional pedestrian-bicycle mixed flows. Furthermore, the calibrated model was applied to evaluate the conflict risk of pedestrians and bicycles in a straight road and an intersection on campus. The safety assessment results indicate that lower density and fewer bicycles in the mixed flow can effectively reduce the risk of conflicts. This study can help understand interactions of pedestrians and cyclists in mixed flow conditions and provide theoretical support for the planning and safety evaluation of pedestrian-bicycle shared roads.

Analysis of the Interaction between Humans and Autonomous Vehicles Equipped with External Human–Machine Interfaces: The Effect of an Experimental Reward Mechanism on Pedestrian Crossing Behavior in a Virtual Environment

The advent of autonomous vehicles (AVs) has sparked many concerns about pedestrian safety, prompting manufacturers and researchers to integrate external Human–Machine Interfaces (eHMIs) into AVs as communication tools between vehicles and pedestrians. The evolving dynamics of vehicle–pedestrian interactions make eHMIs a compelling strategy for enhancing safety. This study aimed to examine the contribution of eHMIs to safety while exploring the impact of an incentive system on pedestrian risk behavior. Participants interacted with AVs equipped with eHMIs in an immersive environment featuring two distinct scenarios, each highlighting a sense of urgency to reach their destination. In the first scenario, participants behaved naturally without specific instructions, while in the second scenario, they were informed of an incentive aimed at motivating them to cross the road promptly. This innovative experimental approach explored whether motivated participants could maintain focus and accurately perceive genuine risk within virtual environments. The introduction of a reward system significantly increased road-crossings, particularly when the vehicle was approaching at higher speeds, indicating that incentives encouraged participants to take more risks while crossing. Additionally, eHMIs notably impacted pedestrian risk behavior, with participants more likely to cross when the vehicle signaled it would not stop.

A Unified Environmental Network for Pedestrian Trajectory Prediction

Accurately predicting pedestrian movements in complex environments is challenging due to social interactions, scene constraints, and pedestrians' multimodal behaviors. Sequential models like long short-term memory fail to effectively integrate scene features to make predicted trajectories comply with scene constraints due to disparate feature modalities of scene and trajectory. Though existing convolution neural network (CNN) models can extract scene features, they are ineffective in mapping these features into scene constraints for pedestrians and struggle to model pedestrian interactions due to the loss of target pedestrian information. To address these issues, we propose a unified environmental network based on CNN for pedestrian trajectory prediction. We introduce a polar-based method to reflect the distance and direction relationship between any position in the environment and the target pedestrian. This enables us to simultaneously model scene constraints and pedestrian social interactions in the form of feature maps. Additionally, we capture essential local features in the feature map, characterizing potential multimodal movements of pedestrians at each time step to prevent redundant predicted trajectories. We verify the performance of our proposed model on four trajectory prediction datasets, encompassing both short-term and long-term predictions. The experimental results demonstrate the superiority of our approach over existing methods.

Autonomous Vehicle and Pedestrian Interaction

Driving assistance systems and even autonomous driving have and will have an important role in sustainable mobility systems. Traffic situations where participants’ cognitive levels are different will cause challenges in the long term. When a pedestrian crosses the road, an autonomous vehicle may need to navigate safely while maintaining its desired speed. Achieving this involves using a predictive model to anticipate pedestrian movements and a strategy for the vehicle to adjust its speed proactively. This research combined model-based predictive control (MPC) with a social-force model (SFM) to effectively control the autonomous vehicle’s longitudinal speed. A genetic algorithm (GA) was also integrated into the approach to address the optimisation problem. A comparison between the proposed approach (MPC-GA) and the conventional MPC technique proved the outperformance of MPC-GA.

Crossing roads in a social context: How behaviors of others shape pedestrian interaction with automated vehicles

Automated vehicles (AVs) are going to enter public roads in the near future and will inevitably encounter more than one pedestrian on roads. Very little is known about how multiple pedestrians will interact with AVs and their external human–machine interfaces (eHMIs). This study investigates the social influence of others on the pedestrian decision-making process during their initial encounters with AVs. In a virtual reality experiment, 26 male and 55 female participants in three experimental conditions made crossing decisions – either alone or next to other simulated pedestrians – in front of AVs equipped with abstract eHMIs showing yielding or non-yielding intentions. The results show that the behaviors of others significantly influenced pedestrian crossing decisions. While the appropriate behaviors of others contributed to faster crossing decision-making, misleading behaviors of others led to riskier crossing decisions and greater difficulty in understanding the eHMIs. However, the impacts from others on pedestrian crossing decisions disappeared once participants became more familiar with eHMIs. Qualitative data revealed that participants acquired knowledge about AVs and eHMIs through direct experience, observing others, and drawing from their past experiences. Additionally, there was a tendency among certain pedestrians to overly rely on and misuse eHMIs. To ensure the safe deployment of AVs, it is therefore essential to adequately educating the public about this novel technology and to optimize eHMI designs for effective communication with multiple pedestrians.