Ego Vehicle Speed Research Articles

Ego-Vehicle Speed Correction for Automotive Radar Systems Using Convolutional Neural Networks.

The development of autonomous driving vehicles has increased the global demand for robust and efficient automotive radar systems. This study proposes an automotive radar-based ego-vehicle speed detection network (AVSD Net) model using convolutional neural networks for estimating the speed of the ego vehicle. The preprocessing and postprocessing methods used for vehicle speed correction are presented in detail. The AVSD Net model exhibits characteristics that are independent of the angular performance of the radar system and its mounting angle on the vehicle, thereby reducing the loss of the maximum detection range without requiring a downward or wide beam for the elevation angle. The ego-vehicle speed is effectively estimated when the range-velocity spectrum data are input into the model. Moreover, preprocessing and postprocessing facilitate an accurate correction of the ego-vehicle speed while reducing the complexity of the model, enabling its application to embedded systems. The proposed ego-vehicle speed correction method can improve safety in various applications, such as autonomous emergency braking systems, forward collision avoidance assist, adaptive cruise control, rear cross-traffic alert, and blind spot detection systems.

Spatio-temporal deep learning framework for pedestrian intention prediction in urban traffic scenes

Pedestrian intent prediction is an essential task for ensuring the safety of pedestrians and vehicles on the road. This task involves predicting whether a pedestrian intends to cross a road or not based on their behavior and surrounding environment. Previous studies have explored feature-based machine learning and vision-based deep learning models for this task but these methods have limitations in capturing the global spatio-temporal context and fusing different features of data effectively. To address these issues, we propose a novel hybrid framework HSTGCN for pedestrian intent prediction that combines spatio-temporal graph convolutional neural networks (STGCN) and long short-term memory (LSTM) networks. The proposed framework utilizes the strengths of both models by fusing multiple features, including skeleton pose, trajectory, height, orientation, and ego-vehicle speed, to predict their intentions accurately. The framework’s performance have been evaluated on the JAAD benchmark dataset and the results show that it outperforms the state-of-the-art methods. The proposed framework has potential applications in developing intelligent transportation systems, autonomous vehicles, and pedestrian safety technologies. The utilization of multiple features can significantly improve the performance of the pedestrian intent prediction task.

Knowledge Distillation from Single-Task Teachers to Multi-Task Student for End-to-End Autonomous Driving

In the domain of end-to-end autonomous driving, conventional sensor fusion techniques exhibit inadequacies, particularly when facing challenging scenarios with numerous dynamic agents. Imitation learning hampers the performance by the expert and encounters issues with out-of-distribution challenges. To overcome these limitations, we propose a transformer-based algorithm designed to fuse diverse representations from RGB-D cameras through knowledge distillation. This approach leverages insights from multi-task teachers to enhance the learning capabilities of single-task students, particularly in a Reinforcement Learning (RL) setting. Our model consists of two primary modules: the perception module, responsible for encoding observation data acquired from RGB-D cameras and performing tasks such as semantic segmentation, semantic depth cloud mapping (SDC), ego vehicle speed estimation, and traffic light state recognition. Subsequently, the control module decodes these features, incorporating additional data, including a rough simulator for static and dynamic environments, to anticipate waypoints within a latent feature space. Vehicular controls (e.g., steering, throttle, and brake) are obtained directly from measurement features and environmental states using the RL agent and are further refined by a PID algorithm that dynamically follows waypoints. The model undergoes rigorous evaluation and comparative analysis on the CARLA simulator across various scenarios, encompassing normal to adversarial conditions. Our code is available at https://github.com/pagand/e2etransfuser/ to facilitate future studies.

Research on car-following control and energy management strategy of hybrid electric vehicles in connected scene

To address the comprehensive optimization problem of driving performance and fuel economy in the driving process of hybrid electric vehicles (HEV) in the car-following scene in the connected environment, an energy management strategy (EMS) based on front vehicle speed prediction and ego vehicle speed planning is designed by combining intelligent transportation system (ITS) technology. The front vehicle speed predictor is first established based on the long short-term memory neural network (LSTM). Then, based on the predicted speed of the front car, the predictive cruise control (PCC) strategy is designed for realizing the speed control in the car-following scene by combining it with the adaptive cruise control (ACC). Finally, based on the planned vehicle speed, deep reinforcement learning (DRL)-based EMS is used to optimize the power distribution among different power components of HEVs. The analysis of simulation results under the SUMO-Python joint simulation platform verifies the proposed strategy.

Adaptive eco-cruising control for connected electric vehicles considering a dynamic preceding vehicle

Energy consumption and driving safety of a vehicle are greatly influenced by the driving behaviors of the vehicle in front (also termed the preceding vehicle). Inappropriate responses to unanticipated changes in the preceding vehicle can lead to decreased energy efficiency and an increased risk of rear-end collisions. To address this issue, this study proposes an innovative Adaptive Eco-cruising Control Strategy (AECS) for connected electric vehicles (CEVs) considering the dynamic behavior prediction of the preceding vehicle. The AECS, which is designed with a two-stage receding horizon control framework, can adapt to scenarios where the preceding vehicle cuts in or moves out in a safer and energy-efficient manner compared to traditional eco-cruising strategies, which merely focus on a constant preceding vehicle. In the first stage, a prediction model for characterizing the dynamic behavior of preceding vehicles is developed using the Bayesian network. This model is trained using real-world vehicle driving data, allowing it to anticipate the driving trajectories of vehicles changing lanes in front. In the second stage, an energy-saving, safety, and driving comfort-oriented optimization problem is formulated as a quadratic programming form. The eco-cruising speed is then optimized to adapt to the dynamic traffic environment, especially when the preceding vehicle changes over time. Finally, several simulations are conducted to validate the AECS. The results demonstrate that the AECS can improve the energy efficiency of CEVs by up to 11.80% and 19.53% on average compared to the existing cruise control strategies and ensure vehicle driving safety and comfort, without compromising travel time. Additionally, the vehicle cut-in position, the cut-in vehicle speed, and the ego vehicle speed affect the energy efficiency improvement performance of the AECS.

Energy Management Strategy Combining Double Deep Q‐Networks and Demand Torque Prediction for Connected Hybrid Electric Vehicles

This paper proposes an energy management strategy (EMS) based on double‐deep Q‐Networks (DDQN) with demand torque prediction (DTP) to optimize the fuel consumption of hybrid electric vehicles (HEVs) by online utilizing vehicle‐to‐vehicle (V2V) information. The main framework of the EMS is designed as DDQN, combining Q‐learning with the deep neural network to realize real‐time training and online optimization of the torque distribution between the internal combustion engine and the electric motor, and given that the model training easily falls into local minima and overestimation, two different Q‐networks are designed to decouple action selection and target evaluation. Meanwhile, BP‐network predicts future demand torque using the ego vehicle speed, front vehicle speed, and distance between the two vehicles to reduce the burden of model training due to the introduction of traffic information. The predicted demand torque is introduced into the DDQN‐based EMS as states, together with the distance traveled in the driving cycle and the current information on HEV. The effectiveness and adaptability to actual driving cycles of the proposed strategy are verified by comparison with DDQN‐based EMS without DTP and rule‐based EMS. The results show that the fuel consumption using DDQN‐based EMS without DTP is reduced by 4.1% compared with rule‐based EMS. The introduction of demand torque prediction resulted in a further 3.9% reduction in fuel consumption. © 2023 Institute of Electrical Engineer of Japan and Wiley Periodicals LLC.

Eco-Driving Optimization of a Signalized Route With Extended Traffic State Information

Literature suggests that driving style and conditions play a major role in vehicle energy consumption. In this sense, this work focuses on vehicle speed planning using information from the environment, through vehicle-to-infrastructure (V2I), and from nearby vehicles, with vehicle-to vehicle (V2V) information to reduce fuel consumption over a signalized route. By knowing the traffic lights scenario of the route in advance and the current position and speed of the preceding vehicle, the proposed algorithm decides the ego-vehicle speed profile during a given horizon to minimize fuel consumption. The proposed strategy solves the optimal control problem (OCP) in each prediction horizon through dynamic programming (DP) with a simplified model. The scenario and the optimal solution are updated periodically to make up for scenario prediction and modeling uncertainties. Experimental tests were conducted on a test bench to evaluate the fuel consumption of the simulated speed profile when compared to the preceding vehicle. Results show that a reduction of almost 20% in fuel consumption is possible without penalizing travel time while keeping it real-time (RT) feasible.

Hierarchical speed planning and energy management for autonomous plug-in hybrid electric vehicle in vehicle-following environment

In this paper, a hierarchical energy management control strategy is investigated for autonomous plug-in hybrid electric vehicle in vehicle-following environment. With the target of safety and comfort, the designed algorithm is divided into two layers. The grey neural network is leveraged in the upper layer controller to predict the future speed trend of preceding vehicle, and the target speed of ego vehicle is planned by fuzzy adaptive control algorithm. By combining with the planned state of charge reference trajectory, genetic algorithm is exploited in the adaptive equivalent consumption minimization strategy-based lower layer controller to determine the initial equivalent factor map by offline iterative calculation, and the fuzzy logic algorithm is employed to update the equivalent factor in real time according to the state of charge difference. Finally, the simulation and hardware-in-the-loop experiment are conducted to validate the performance of the proposed strategy. The simulation results highlight the capability of the proposed strategy in solving multi-objective optimization for autonomous plug-in hybrid electric vehicle in vehicle-following environment, and the experiment results validate that the energy consumption economy of the proposed strategy reaches 95.43% optimality of the results derived by dynamic programming while ensuring the satisfied driving comfort and safety.

Predicting Pedestrian Crossing Intention With Feature Fusion and Spatio-Temporal Attention

Predicting vulnerableroad user behavior is an essential prerequisite for deploying Automated Driving Systems (ADS) in the real-world. Pedestrian crossing intention should be recognized in real-time, especially for urban driving. Recent works have shown the potential of using vision-based deep neural network models for this task. However, these models are not robust and certain issues still need to be resolved. First, the global spatio-temporal context that accounts for the interaction between the target pedestrian and the scene has not been properly utilized. Second, the optimal strategy for fusing different sensor data has not been thoroughly investigated. This work addresses the above limitations by introducing a novel neural network architecture to fuse inherently different spatio-temporal features for pedestrian crossing intention prediction. We fuse different phenomena such as sequences of RGB imagery, semantic segmentation masks, and ego-vehicle speed in an optimal way using attention mechanisms and a stack of recurrent neural networks. The optimal architecture was obtained through exhaustive ablation and comparison studies. Extensive comparative experiments on the JAAD and PIE pedestrian action prediction benchmarks demonstrate the effectiveness of the proposed method, where state-of-the-art performance was achieved. Our code is open-source and publicly available: <uri>https://github.com/OSU-Haolin/Pedestrian_Crossing_Intention_Prediction</uri>.

合流路での車列間に入るための車速制御方法

A method of controlling the acceleration required for merging between the vehicles on a main road was investigated. The vehicles on the main road usually keep speed near junctions. In order to prevent collisions, at the time of merging, the speed of ego vehicle, and the inter-vehicle distance of object vehicles on the main road and the distance to the merging point are measured previously. Using these parameters, we derived a calculation method of the acceleration required for safe merging between the vehicles on a main road. In addition, as the acceptable upper and lower limits of the vehicle's acceleration are defined. It is possible to predict whether the vehicle can safely merge with acceptable acceleration/deceleration.

Desired Relative Distance Model-based Personalized Braking Algorithm for One-pedal Driving of Electric Vehicles

One-pedal driving (OPD) provides a unique experience to a driver during car-following as an important feature for energy saving in electric vehicles. However, the braking behavior of one driver may differ significantly from that of another; therefore, it is necessary to develop a personalized braking algorithm to reduce unpleasantness for the passengers. In addition, for ease of implementation in a real vehicle, a relatively simple structure is preferred. This paper proposes a desired relative distance-based personalized braking (DRD-PB) algorithm to achieve these requirements. The desired relative distance model is formulated as a function of the ego vehicle speed and the relative speed in a braking scenario in which a preceding vehicle exists. The proposed algorithm is calibrated and validated against single driver-driven data collected in various driving conditions. The simulation results show that the proposed algorithm can achieve satisfactory braking performance close to that of a human driver: the median of the root mean square errors of drel, vego, and aego are 4.3 m, 0.95 m/s, and 0.46 m/s2, respectively.

CAPformer: Pedestrian Crossing Action Prediction Using Transformer.

Anticipating pedestrian crossing behavior in urban scenarios is a challenging task for autonomous vehicles. Early this year, a benchmark comprising JAAD and PIE datasets have been released. In the benchmark, several state-of-the-art methods have been ranked. However, most of the ranked temporal models rely on recurrent architectures. In our case, we propose, as far as we are concerned, the first self-attention alternative, based on transformer architecture, which has had enormous success in natural language processing (NLP) and recently in computer vision. Our architecture is composed of various branches which fuse video and kinematic data. The video branch is based on two possible architectures: RubiksNet and TimeSformer. The kinematic branch is based on different configurations of transformer encoder. Several experiments have been performed mainly focusing on pre-processing input data, highlighting problems with two kinematic data sources: pose keypoints and ego-vehicle speed. Our proposed model results are comparable to PCPA, the best performing model in the benchmark reaching an F1 Score of nearly against . Furthermore, by using only bounding box coordinates and image data, our model surpasses PCPA by a larger margin ( vs. ). Our model has proven to be a valid alternative to recurrent architectures, providing advantages such as parallelization and whole sequence processing, learning relationships between samples not possible with recurrent architectures.

Analysis of drivers’ deceleration behavior based on naturalistic driving data

Objective: As one of the bases for designing a humanlike brake control system for the intelligent vehicle, drivers’ deceleration behavior needs to be understood. There are two modes for drivers’ deceleration behavior: (i) brake pedal input, by applying brake system to reduce the speed; (ii) no pedal input, by releasing the accelerator pedal without pressing the brake pedal, thus decelerating by naturalistic driving resistance. The deceleration behavior that drivers choose to press the brake pedal has been investigated in previous studies. However, releasing the accelerator pedal behavior has not received much attention. The objective of this study is to investigate factors that influence drivers’ choice of the two deceleration modes using naturalistic driving data, which provide a theoretical foundation for the design of the brake control system.Methods: A logistic model was constructed to model drivers’ deceleration mode, valued as “no pedal input” or “brake pedal input” for dependent variables. Factors such as Light condition, Intersection mode, Road alignment, Traffic flow, Traffic light, Ego-vehicle motion state, Lead vehicle motion state, Time headway (THW), and Ego-vehicle speed were considered in the model as independent variables.Results: 393 deceleration events were selected from the naturalistic driving data, which used as the database for the regression model. As a result, 6 remarkable factors were found to influence drivers’ deceleration model, which include Traffic flow, Intersection mode, Lead vehicle motion state, Ego-vehicle motion state, Ego-vehicle speed and THW. Specifically, (1) the possibility of drivers choosing “no pedal input” is gradually increasing with the increase of THW and speed; (2) The drivers prefer to choose “no pedal input” when the lead vehicle is decelerating compared to it’s stationary. This probability is relatively high when the lead vehicle is traveling along the road; (3) the possibility of choosing “no pedal input” at intersection is higher than roads without intersection; (4) the possibility of choosing “no pedal input” is higher when traveling with more traffic flow.Conclusion: The drivers’ deceleration behavior can be divided into “no pedal input” and “brake pedal input.” The following six factors significantly affect drivers’ choice of deceleration mode: Traffic flow, Intersection mode, Lead vehicle motion state, Ego-vehicle motion state, Ego-vehicle speed and THW. The logistic regression model can quantify the influence of these six factors on drivers’ deceleration behavior. This study provides a theoretical basis for the braking system design of ADAS (Advanced Driving Assistant System) and intelligent control system.

Ego-Vehicle Speed Prediction Using a Long Short-Term Memory Based Recurrent Neural Network

for predictive powertrain control, accurate prediction of vehicle speed is required. As vehicle speed prediction is affected by the driver’s response to numerous driving conditions under uncertainty, the development of an accurate model is quite challenging. This paper proposes an ego-vehicle speed prediction model using a long short-term memory (LSTM) based recurrent neural network (RNN). The proposed model uses various inputs to increase the prediction accuracy: internal vehicle information, relative speed and distance to the vehicle ahead measured by a radar sensor, and the ego-vehicle location estimated by the GPS signal and B-spline roadway model. The LSTM based RNN model predicts the ego-vehicle speed for 15 seconds by using inputs from the past 30 seconds. The model was evaluated by real driving data for three scenarios: car-following, sharp curve road, and full path. In all scenarios, the radar sensor and the information of the location of the ego-vehicle contribute to improvement of the speed prediction accuracy. Thus, we conclude that for application of the predictive powertrain control, besides the internal vehicle information, the radar sensor, and the location of the ego-vehicle information are critical inputs to the speed prediction model.

Design of Conditional Driving Automation Variables to Improve Takeover Performance

Abstract Conditional driving automation occasionally decouples drivers from driving to participate in non-driving related tasks. On the other hand, situation awareness of drivers in some driving situations that can not be managed by the automated system is required to avoid accidents. In recent years, it has been observed that surrounding traffic conditions, complexity of the driving scenario, secondary tasks, speed of ego vehicle, and takeover request experience affect takeover performance. However, the scope of influence of the variables is not known in detail. This contribution discusses an approach for systematically designing these variables. The approach utilizes known dimensions of complex dynamical systems and fundamentals of multitasking to design driving scenarios and non-driving related tasks. The effect of the variables are systemically varied to generate different driving situations to better understand their scope and interaction.

Smart and Green Adaptive Cruise Control for an Electric Vehicle: first results

This paper concerns the development of the Smart and Green Adaptive Cruise Control aiming at controlling the ego vehicle speed taking into account the road data in order to optimize the energy consumption while improving safety (no collision, under the legal speed, up to the safe headway). A graph search algorithm of Dijkstra is used to build up a velocity pattern offering the best compromise between energy consumed and covered time.