Maneuver Prediction Research Articles

RAG-based explainable prediction of road users behaviors for automated driving using knowledge graphs and large language models

The prediction of road user behaviors in the context of autonomous driving has attracted considerable attention from the scientific community in recent years. Most works focus on predicting behaviors based on kinematic information alone, a simplification of reality since road users are humans, and as such they are highly influenced by their surrounding context. In addition, a large plethora of research works rely on powerful Deep Learning techniques, which exhibit high-performance metrics in prediction tasks but may lack the ability to fully understand and exploit the contextual semantic information contained in the road scene, not to mention their inability to provide explainable predictions that can be understood by humans. In this work, we propose an explainable road users’ behavior prediction system that integrates the reasoning abilities of Knowledge Graphs (KG) and the expressiveness capabilities of Large Language Models (LLM) by using Retrieval Augmented Generation (RAG) techniques. For that purpose, Knowledge Graph Embeddings (KGE) and Bayesian inference are combined to allow the deployment of a fully inductive reasoning system that enables the issuing of predictions that rely on legacy information contained in the graph, as well as on current evidence gathered in real-time by onboard sensors. Two use cases have been implemented following the proposed approach: 1) Prediction of pedestrians’ crossing actions; and 2) Prediction of lane change maneuvers. In both cases, the performance attained exceeds the current state-of-the-art in terms of anticipation and F1 score, showing a promising avenue for future research in this field.

Failed lane-changing detection and prediction using naturalistic vehicle trajectories

Lane-changing maneuvers are crucial driving behaviors closely linked to various collisions, such as rear-end and sideswipe collisions. Precisely predicting lane-changing maneuvers can aid drivers in making informed decisions, thus enhancing driving safety. However, existing research primarily focuses on successful lane-changing maneuvers, neglecting failed ones. This study comprehensively investigates the detection and prediction of failed lane-changing maneuvers in discretionary scenarios using naturalistic vehicle trajectory data. The Mexican hat wavelet (MHW) is employed to accurately detect key time points in successful or failed lane-changing events, including the start, occurrence of failure, and end of the lane-changing maneuver. Subsequently, a failed lane-changing prediction model based on GA-XGBoost is developed to proactively perceive whether a lane-changing maneuver will succeed before it initiates. Moreover, the Shapley Additive exPlanations (SHAP) technique assesses feature importance and interaction effects between features in the prediction model. The results demonstrate that MHW effectively identifies critical time points in lane-changing events. The proposed GA-XGBoost model achieves an impressive 94.55% accuracy in predicting failed lane-changing maneuvers before the driver initiates the lane-changing maneuver. SHAP values reveal that failed lane-changing maneuvers often result from a high collision risk between the subject vehicle (SV) and the following vehicle in the target lane (FVT). Consequently, drivers should pay increased attention to approaching vehicles from behind. Moreover, accelerating during lane-changing helps maintain a safe distance from FVT, improving the likelihood of a successful lane-changing. Integrating the proposed model into advanced driver-assistance systems or autonomous driving systems has the potential to significantly enhance driving safety.

Novel hawk swarm‐optimized deep learning classification with K‐nearest neighbor based decision making for autonomous vehicle movement controller

SummaryNowadays, intelligent transportation systems pay a lot of attention to autonomous vehicles it is believed that an autonomous vehicle improves mobility, comfort, safety, and energy efficiency. Making decisions is essential for the development of autonomous vehicles since these algorithms must be able to manage dynamic and complex urban crossings. In this research an optimal deep BiLSTM‐GAN classifier to detect the movement of smart vehicles, initially the preprocessing stage is performed to decrease noise in the received data after that the essential regions are next be extracted in the region of interest (ROI) to make the right decision. The extracted data are forwarded to the GAN for road segmentation as well as the optimized deep BiLSTM classifier, which recognizes the traffic sign, simultaneously making it possible to do a modified Hough line‐based maneuver prediction using the segmented information from the roads. Finally, the GAN determines the lane, and the BiLSTM predicts the traffic sign. The K‐nearest neighbor (KNN)‐based autonomous vehicle movement controllers are used to make the decision based on the predicted traffic sign and information about the lane. The proposed HSO algorithm was developed as the outcome of the common fusion of hawk and swarm optimization. Based on lane detecting achievements, at training percentage (TP) 90, the accuracy is 91.75%, Peak signal‐to‐noise ratio (PSNR) is 64.84%, mean square error (MSE) is 28.78, and mean absolute error (MAE) is 20.20, respectively, similarly based on the traffic sign prediction achievements at TP 90, the accuracy is 93.71%, sensitivity is 95.15%, specificity is 93.91%, and MSE is 28.78%, respectively.

Determination of Propeller-Rudder-Hull Interaction Coefficients in Ship Manoeuvring Prediction

Abstract The assessment of ship manoeuvring properties is a crucial part of the process of ship design and is usually first carried out during the model test phase of the project. According to the International Maritime Organisation (IMO), the manoeuvrability of the ship can be assessed on the basis of the standard trial manoeuvres. In order to do this, free running model tests or captive model tests are used, in conjunction with a mathematical model of ship motion; this is considered to be a reliable prediction method. In recent years, numerical-based methods have also been widely used in ship hydrodynamics and constantly improving computing power and more accurate fluid dynamics models have made the simulation of more complex cases possible. The study presented in this paper focuses on the determination of propeller-rudder-hull interaction coefficients based on the Mathematical Modelling Group (MMG) standard method in ship manoeuvring prediction. The identification of the parameters uses both captive model tests and a simplified numerical method, as well as regression formulas. The results of 35° turning and 10°/10° zig-zag manoeuvres, obtained with the use of each prediction method, are then compared. The test case used in the study is the container type cargo ship equipped with a single propeller and rudder. The model scale, for which the referenced model tests were carried out, is equal to 1:25 and a NACA 0020 rudder profile was used. This research highlights the advantages and disadvantages of each presented prediction method and their potential for future improvement.

Numerical simulation of near-surface turn maneuver of a submarine in waves

Aiming toward a better understanding of the limitations of submarine maneuverability in complex marine environments, systematic investigations of near-surface maneuvers of a generic submarine in waves were conducted. By using the sliding grid for propeller rotation and the overset grid for control plane deflection, numerical simulations of the free running submarine model under different near-surface maneuver conditions were performed. The fifth-order Stokes wave was utilized to model the regular wave. Simulated results of near-surface self-propulsion and turn maneuvers with different wavelengths, wave heights, and submergence depths were analyzed. The results demonstrated that in near-surface turn maneuvers in waves, to balance the dramatic fluctuation of the submarine's attitude, a significant dynamic deflection of rudders with the maximum angle probably reaching the upper limit was observed. The resultant effectiveness of rudders varied greatly, which directly led to a significant difference in the trajectories. The research can provide more comprehensive guidance for the prediction and evaluation of submarine maneuverability and seaworthiness.

Integrating Computational Fluid Dynamics for Maneuverability Prediction in Dual Full Rotary Propulsion Ships: A 4-DOF Mathematical Model Approach

To predict the maneuverability of a dual full rotary propulsion ship quickly and accurately, the integrated computational fluid dynamics (CFD) and mathematical model approach is performed to simulate the ship turning and zigzag tests, which are then compared and validated against a full-scale trial carried out under actual sea conditions. Initially, the RANS equations are solved, employing the Volume of Fluid (VOF) method to capture the free water surface, while a numerical simulation of the captive model test is conducted using the rigid body motion module. Secondly, hydrodynamic derivatives for the MMG model are obtained from the CFD simulations and empirical formula. Lastly, a four-degree-of-freedom mathematical model group (MMG) maneuvering model is proposed for the dual full rotary propulsion ship, incorporating full-scale simulations of turning and zigzag tests followed by a full-scale trial for comparative validation. The results indicate that the proposed method has a high accuracy in predicting the maneuverability of dual full-rotary propulsion ships, with an average error of less than 10% from the full-scale trial data (and within 5% for the tactical diameters in particular) in spite of the influence of environmental factors such as wind and waves. It provides experience in predicting the maneuverability of a full-scale ship during the ship design stage.

A review of driver gaze estimation and application in gaze behavior understanding

Driver gaze plays a key role in different gaze-based applications, such as driver attentiveness detection, visual distraction detection, gaze behavior understanding, and building driver assistance system. The main objective of this study is to perform a comprehensive summary of driver gaze fundamentals, methods to estimate driver gaze using machine learning (ML) based technique, and its applications in real world driving scenarios. We first discuss the fundamentals related to driver gaze, involving head-mounted and remote setup based gaze estimation and the terminologies used for driver gaze behavior understanding. Next, we list out the existing benchmark driver gaze datasets, highlighting the collection methodology, and the equipment used for such data collection. This is followed by a discussion of the algorithms used for driver gaze estimation, which primarily involves traditional machine learning and deep learning (DL) based techniques. The estimated driver gaze is then used for understanding gaze behavior while maneuvering through intersections, on-ramps, off-ramps, lane changing, determining the effect of roadside advertising structures and also for developing driver gaze based applications such as maneuver prediction, driver inattention, and distraction detection systems, etc. Finally, we discuss the limitations in the existing literature, challenges, and future scope in driver gaze estimation and gaze-based applications.

Fully Memristive Elementary Motion Detectors for a Maneuver Prediction.

Insects can efficiently perform object motion detection via a specialized neural circuit, called an elementary motion detector (EMD). In contrast, conventional machine vision systems require significant computational resources for dynamic motion processing. Here, a fully memristive EMD (M-EMD) is presented that implements the Hassenstein-Reichardt (HR) correlator, a biological model of the EMD. The M-EMD consists of a simple Wye (Y) configuration, including a static resistor, a dynamic memristor, and a Mott memristor. The resistor and dynamic memristor introduce different signal delays, enabling spatio-temporal signal integration in the subsequent Mott memristor, resulting in a direction-selective response. In addition, a neuromorphic system is developed employing the M-EMDs to predict a lane-changing maneuver by vehicles on the road. The system achieved a high accuracy (> 87%) in predicting future lane-changing maneuvers on the Next Generation Simulation (NGSIM) dataset while reducing the computational cost by 92.9% compared to the conventional neuromorphic system without the M-EMD, suggesting its strong potential for edge-level computing.

Mechanics of vaginal breech birth: Factors influencing obstetric maneuver rate, duration of active second stage of labor, and neonatal outcome.

We investigated possible parameters that could predict the need for obstetric maneuvers, the duration of the active second stage of labor (i.e., the duration of active pushing), and short-term neonatal outcome in vaginal breech births. We performed a retrospective analysis of 268 successful singleton vaginal breech births in women without previous vaginal births from January 2015 to August 2022. Multivariable regression was used to investigate associations between maternal and fetal characteristics (including antepartum magnetic resonance (MR) pelvimetry) with obstetric maneuvers, the duration of active second stage of labor, pH values, and admission to the neonatal unit. Models for the prediction of obstetric maneuvers were built and internally validated. Obstetric maneuvers were performed in a total of 130 women (48.5%). A total of 32 neonates (11.9%) had to be admitted to the neonatal unit. The intertuberous distance (ITD) (p < 0.001), epidural analgesia (p < 0.001), and birthweight (p = 0.026) were associated with the duration of active second stage of labor. ITD (p = 0.028) and birthweight (p = 0.011) were also independently associated with admission to the neonatal unit, while pH values below 7.10 dropped significantly (p = 0.0034) if ITD was ≥13 cm. Furthermore, ITD (p < 0.001) and biparietal diameter (p = 0.002) were independent predictors for obstetric maneuvers. ITD is independently associated with the duration of active second stage of labor. Thus, it can predict suboptimal birth mechanics in the last stage of birth, which may lead to the need for obstetric maneuvers, lower arterial pH values, and admission to the neonatal unit. Consequently, MR pelvimetry gives additional information for practitioners and birthing people preferring a vaginal breech birth.

Estimating Hydrodynamic Coefficients of Real Ships Using AIS Data and Support Vector Regression

In response to the complexity and time demands of conventional methods for estimating the hydrodynamic coefficients, this study aims to revolutionize ship maneuvering analysis by utilizing automatic identification system (AIS) data and the Support Vector Regression (SVR) algorithm. The AIS data were collected and processed to remove outliers and impute missing values. The rate of turn (ROT), speed over ground (SOG), course over ground (COG) and heading (HDG) in AIS data were used to calculate the rudder angle and ship velocity components, which were then used as training data for a regression model. The accuracy and efficiency of the algorithm were validated by comparing SVR-based estimated hydrodynamic coefficients and the original hydrodynamic coefficients of the Mariner class vessel. The validated SVR algorithm was then applied to estimate the hydrodynamic coefficients for real ships using AIS data. The turning circle test wassimulated from calculated hydrodynamic coefficients and compared with the AIS data. The research results demonstrate the effectiveness of the SVR model in accurately estimating the hydrodynamic coefficients from the AIS data. In conclusion, this study proposes the viability of employing SVR model and AIS data for accurately estimating the hydrodynamic coefficients. It offers a practical approach to ship maneuvering prediction and control in the maritime industry.

Self-organizing data-driven prediction model of ship maneuvering fast-dynamics

In this paper, to facilitate ship maneuvering fast-dynamics prediction which is critically required within motion planning and control, a self-organizing data-driven network with hierarchical pruning (SDN-HP) is innovated by virtue of fuzzy neural architecture. To be specific, the SDN-HP model is incrementally trained from an empty-hidden-node fuzzy neural network with recurrent velocities as inputs and outputs, and the rudder angle as exogenous command. With the aid of fuzzy neuron generation criteria, the SDN grows with increasing fuzzy rules such that fast-dynamics of ship velocities can be in-time fitted. To balance training accuracy and prediction generalization, hierarchical pruning mechanism is created by defining a new concept of fuzzy rule intensity, in order that hidden nodes with insignificant contributions can be dynamically removed, thereby enabling dynamics abstraction with parsimonious structure. With the aid of reference model of Esso Osaka tanker and KVLCC2 database, the SDN-HP model is sufficiently trained and applied to free-running maneuvering prediction in on-line manners. Experimental results show that the proposed SDN-HP framework achieves remarkable fast-dynamics prediction ability and outperforms state-of-the-art deep learning paradigm, e.g., LSTM and GRU.

Numerical simulation of near-surface turn maneuver of a generic submarine

Aiming at the actual demand of improving the numerical simulation of the unsteady motion of submarines in maneuvering conditions, the overset grid is used to model six degrees of freedom coupling motions of the submarine and the independent motion of each control plane, and the sliding grid is used to simulate the rotation of the propeller, to carry out the research on the numerical simulation approach of a generic submarine in near-surface maneuvering condition. The VOF method is utilized to capture the free surface. By comparing the submerged turn maneuver numerical simulation results of typical dynamics parameters with the experiment's results, the practicability of the approach for engineering prediction is verified. Then the simulations of near-surface maneuvers under different submergence depths are conducted, and the influence of submergence depth on kinematics and dynamics parameters of Joubert BB2 in maneuvering conditions are analyzed. Results demonstrate that the control performance of the stern planes will be sharply degraded when maneuvering too close to the free surface, resulting in a significant increase in the tactical diameter of the turn maneuver. The research can guide the prediction and evaluation of submarine maneuverability and seaworthiness. It is beneficial for the improvement of free running submarine model tests.

Maneuver Prediction Using Traffic Scene Graphs via Graph Neural Networks and Recurrent Neural Networks

The driving process involves many layers of planning and navigation, in order to enable tractable solutions for the otherwise highly complex problem of autonomous driving. One such layer involves an inherent discrete layer of decision-making corresponding to tactical maneuvers. Inspired by this, the focus of this work is predicting high-level maneuvers for the ego-vehicle. As maneuver prediction is fundamentally feedback-structured, it requires modeling techniques that take into consideration the interaction awareness of the traffic agents involved. This work addresses this challenge by modeling the traffic scenario as an interaction graph and proposing three deep learning architectures for interaction-aware tactical maneuver prediction of the ego-vehicle. These architectures are based on graph neural networks (GNNs) for extracting spatial features among traffic agents and recurrent neural networks (RNNs) for extracting dynamic motion patterns of surrounding agents. These proposed architectures have been trained and evaluated using BLVD dataset. Moreover, this dataset is expanded using data augmentation, data oversampling and data undersampling approaches, to strengthen model’s resilience and enhance the learning process. Lastly, we compare proposed learning architectures for ego-vehicle maneuver prediction in various driving circumstances with various numbers of surrounding traffic agents in order to effectively verify the proposed architectures.

A self-error corrector integrating K-means clustering with Markov model for marine craft maneuvering prediction with experimental verification

A novel approach for maneuvering prediction of marine crafts in data-limited situations is presented in this study. The approach is a hybrid strategy that combines least square-support vector machines (LS-SVM) and a self-error corrector that integrates K-means clustering with Markov model. The strategy allows testing errors caused by LS-SVM to be compensated effectively by the dedicated self-error corrector. To classify the inputs of the Markov model-based error corrector rapidly and properly, the K-means clustering algorithm is resorted to classifying the training errors obtained from LS-SVM. Assuming that maneuvering prediction for marine crafts with a limited number of simulation or experiment data, a comparative analysis of the prediction performance between the proposed strategy and the conventional method based on LS-SVM is performed. The results of numerical simulations and experiments demonstrate the superiority of the proposed strategy, which reduces the root mean square error of surge speed by up to 86.2% in simulation tests and 54.5% in experiments.

An advanced deep learning model for maneuver prediction in real-time systems using alarming-based hunting optimization

The increasing trend of autonomous driving vehicles in smart cities emphasizes the need for safe travel. However, the presence of obstacles, potholes, and complex road environments, such as poor illumination and occlusion, can cause blurred road images that may impact the accuracy of maneuver prediction in visual perception systems. To address these challenges, a novel ensemble model named ABHO-based deep CNN-BiLSTM has been proposed for traffic sign detection. This model combines a hybrid convolutional neural network (CNN) and bidirectional long short-term memory (BiLSTM) with the alarming-based hunting optimization (ABHO) algorithm to improve maneuver prediction accuracy. Additionally, a modified hough-enabled lane generative adversarial network (ABHO based HoughGAN) has been proposed, which is designed to be robust to blurred images. The ABHO algorithm, inspired by the defending and social characteristics of starling birds and Canis kojot, allows the model to efficiently search for the optimal solution from the available solutions in the search space. The proposed ensemble model has shown significantly improved accuracy, sensitivity, and specificity in maneuver prediction compared to previously utilized methods, with minimal error during lane detection. Overall, the proposed ensemble model addresses the challenges faced by autonomous driving vehicles in complex and obstructed road environments, offering a promising solution for enhancing safety and reliability in smart cities.

Adaptive event-triggered mechanism-based online system identification framework for marine craft

Online dynamic model identification is a promising technology for designing the model-based controller of marine crafts. This study confronts the online model identification and maneuvering prediction for marine crafts with an adaptive event-triggered mechanism. In particular, the least square-twin support vector machines algorithm is exploited to identify the dynamic model accurately. Meanwhile, an online adaptive event-triggered identification and prediction framework is proposed to deal with the issue of when to update the dynamic model. Finally, a comparative analysis of the identification and prediction performance between the proposed framework and the non-updated model method is performed. The former superiority is highlighted by experiments based on a marine craft, which shows that it reduces the root mean square error of the yaw rate by up to 45.9%. Thereby, the proposed online identification framework offers great potential for future practical applications of marine crafts.

Towards the effects of load condition on ship manoeuvrability by a system simulation method with hydrodynamic derivatives from RANS computations

The effects of load condition on ship manoeuvrability are investigated by means of system simulation method. To this end, a MMG-type mathematical manoeuvring model is applied to simulate the standard turning and zigzag rudder manoeuvres for a KVLCC2 ship model in six load conditions (two drafts × three trims). In order to determine the needed hydrodynamic derivatives, the forces and moments on the hull in a range of combined static drift and circle motions are computed by using the own expanded RANS (Reynolds-Averaged Navier-Stokes) solver on OpenFOAM. However, the coefficients of hydrodynamic interaction among hull, propeller and rudder are obtained from available experiments. The computed surge forces, sway forces and yaw moments agree fairly well with the experimental data, suggesting a promising prospect for the used RANS solver in the application of ship manoeuvring prediction. By regressing the hydrodynamic forces and moments, the hydrodynamic derivatives are obtained. Turning and zigzag rudder manoeuvres are simulated by solving ship motion equations in line with the acquired derivatives and coefficients. Comparing the simulated results between the full and ballast load condition without trim, a small draft (i.e. the ballast load condition) improves turning performance, but weaken course stability. For the three full load conditions or ballast load conditions, trim by bow improves turning ability, but deteriorates yaw-checking ability. The simulated results for two load conditions show satisfactory agreement with available experimental data, which suggests that the present procedure for ship manoeuvring prediction is effective. The course stability indexes are analysed as well. The load condition will affect significantly the manoeuvrability of a ship, at least for KVLCC2. It may be quite necessary to evaluate the manoeuvrability of a ship in different load conditions at the initial design stage.

Maneuver Conditioned Vehicle Trajectory Prediction Using Self-Attention

Forecasting the motion of surrounding vehicles is necessary for a self-driving vehicle to plan a safe and efficient trajectory for the future. Like experienced human drivers, the self-driving vehicle needs to perceive the interaction of surrounding vehicles and decide the best trajectory from many choices. However, previous methods either lack modeling of interactions or ignore the multi-modal nature of this problem. In this paper, we focus on two important cues of trajectory prediction: interaction and maneuver, and propose Maneuver conditioned Attentional Network named MAN. MAN learns the interactions of all vehicles in a scenario in parallel by self-attention social pooling and the attentional decoder generates the future trajectory conditioned on the predicted maneuver among 3 classes: Lane Changing Left (LCL), Lane Changing Right (LCR) and Lane Keeping (LK). Experiments demonstrate the improvement of our model in prediction on the publicly available NGSIM and HighD datasets. We also present quantitative analysis to study the relationship between maneuver prediction accuracy and trajectory error.

UAV's air combat decision-making based on deep deterministic policy gradient and prediction

To solve the enemy uncertain manipulation problem during a UAV's autonomous air combat maneuver decision-making, this paper proposes an autonomous air combat maneuver decision-making method that combines target maneuver command prediction with the deep deterministic policy algorithm. The situation data of both sides of air combat are effectively fused and processed, the UAV's six-degree-of-freedom model and maneuver library are built. In air combat, the target generates its corresponding maneuver library instructions through the deep Q network algorithm; at the same time, the UAV on our side gives the target maneuver prediction results through the probabilistic neural network. A deep deterministic policy gradient reinforcement learning method that considers both the situation information of two aircraft and the prediction results of enemy aircraft is proposed, so that the UAV can choose the appropriate maneuver decision according to the current air combat situation. The simulation results show that the method can effectively use the air combat situation information and target maneuver prediction information so that it can improve the effectiveness of the reinforcement learning method for UAV's autonomous air combat decision-making on the premise of ensuring convergence.

Twin-screw ASD tug maneuvering prediction based on integrated CFD and empirical methods

Twin-screw ASD tug maneuvering prediction based on integrated CFD and empirical methods