Road driving optimization method and system for vehicle state and traffic factor decoupling

Abstract

A road driving optimization method and system having a vehicle sensor configured to collect external sensor data, vehicle status data, and communication data; a control module configured to analyze the collected data to detect a trigger event, a state of the target vehicle, an achievable speed range, and an instantaneous traction; a decoupling estimator module configured to analyze the trigger event, a state of the target vehicle, an achievable speed range, and the personalized driver profile to determine a maximum free flow distance and an arrival speed at the free flow distance; a machine learning model configured to predict a speed profile of the host vehicle during approaching the trigger event based on the determined free flow distance, the determined arrival speed at the determined free flow distance, and the instantaneous traction; and a cruise control system configured to implement the predicted speed profile.

Description

Technical Field

[0001] The present disclosure generally relates to vehicles having an autonomous driving system, and more particularly to a method and system for optimizing road driving by decoupling vehicle states and traffic factors. Background Art

[0002] Modern intelligent vehicles have intelligent systems, also known as smart systems, such as advanced driver assistance systems (ADAS) and / or autonomous driving systems (ADS), which are located on the intelligent vehicle to enhance or automate the functions of various vehicle systems. The intelligent system has one or more control modules that communicate with vehicle sensors such as external sensors, internal sensors, and status sensors, as well as various vehicle systems such as steering, acceleration, braking, and safety systems. The control module analyzes the information collected by the vehicle sensors and generates instructions to various vehicle systems for operating the vehicle, which can operate between a partial autonomous driving mode and a full autonomous driving mode.

[0003] An adaptive cruise control (ACC) system may be part of an ADAS or an ADS. The ACC system receives various inputs from vehicle sensors and relies on algorithms to control the longitudinal distance between the host vehicle and a target vehicle traveling immediately ahead of the host vehicle. The ACC system can also rely on algorithms to control the lateral position of the host vehicle within a lane by applying torque to the vehicle steering system to keep the vehicle within the lane. Such algorithms are based on fixed rules that are used for the ACC system of the host vehicle to respond to predetermined actions of the target vehicle and other triggering events. Therefore, current ACC systems have limited or even no ability to optimize the response of the host vehicle to real-world triggering events. A triggering event is a traffic event that may cause the host vehicle to need to adjust its speed profile and / or travel direction.

[0004] Although current ACC systems achieve their goals, there is still a need for a system and method to optimize the performance of vehicles in terms of fuel economy and passenger comfort. Summary of the Invention

[0005] According to several aspects, a method of driving a host vehicle is provided. The method includes receiving data collected by at least one vehicle sensor by a control module; analyzing the collected data by the control module to detect a triggering event and determine an achievable speed range; determining a free flow distance and an arrival speed at the free flow distance based on the detected triggering event and the determined achievable speed range by a decoupling estimator module; inputting the determined free flow distance and the determined arrival speed at the free flow distance into a pre-trained machine learning (ML) model to predict a speed profile to reach the triggering event; and driving the host vehicle towards the triggering event based on the predicted speed profile.

[0006] In another aspect of the present disclosure, the method further includes analyzing the collected data to determine the state of the target vehicle, wherein the state of the target vehicle includes the position of the target vehicle relative to the host vehicle and the speed of the target vehicle; wherein the decoupled estimator module further determines the free flow distance and the arrival speed at the free flow distance based on the determined state of the target vehicle.

[0007] In another aspect of the present disclosure, the method further includes inputting the determined free flow distance and the incremental correction of the speed of reaching the free flow distance into the pre-trained ML model, so as to iteratively predict the speed curve.

[0008] In another aspect of the present disclosure, the method further includes obtaining a personalized operator profile; wherein the decoupled estimator module further determines the maximum free flow distance and the arrival speed at the maximum free flow distance based on the obtained personalized operator profile.

[0009] In another aspect of the present disclosure, the method further includes determining the instantaneous traction force of the host vehicle; and inputting the determined instantaneous traction force together with the determined free flow distance and the arrival speed at the free flow distance into the pre-trained ML model to predict the speed curve of reaching the trigger event.

[0010] In another aspect of the present disclosure, the predicted speed curve minimizes fuel consumption and minimizes the jerk motion of the host vehicle during reaching the trigger event.

[0011] In another aspect of the present disclosure, wherein the adaptive cruise control (ACC) system drives the host vehicle towards the trigger event based on the predicted speed curve.

[0012] In another aspect of the present disclosure, the determined free flow distance is greater than a predetermined free flow distance threshold.

[0013] In another aspect of the present disclosure, the pre-trained machine learning (ML) model is trained on a deep neural network.

[0014] In another aspect of the present disclosure, the trigger event is a dynamic traffic light, wherein the collected data includes the current state of the dynamic traffic light and the time interval until the dynamic traffic light changes to the next state.

[0015] According to several aspects, a tangible, non-transitory, machine-readable medium includes machine-readable instructions that, when executed by a processor, cause the processor to: receive external sensor data and vehicle state data; analyze the external sensor data and vehicle state data to determine a trigger event and determine an achievable speed range; input the determined trigger event and the determined achievable speed range into a decoupled estimator to determine a free-flow distance and determine an arrival speed at the free-flow distance; input the determined free-flow distance and the determined arrival speed at the free-flow distance into a machine learning model to predict a speed curve to reach the trigger event; and drive the host vehicle towards the trigger event according to the predicted speed curve.

[0016] In another aspect of the present disclosure, the tangible, non-transitory, machine-readable medium further includes machine-readable instructions that, when executed by a processor, cause the processor to: receive personal driver profile and communication data; analyze the external sensor data, vehicle state data, and communication data to determine a trigger event, an achievable speed range, and a state of a target vehicle; and input the determined trigger event, the determined achievable speed range, the determined state of the target vehicle, and the personal driver profile into a decoupled estimator to determine a free-flow distance and determine an arrival speed at the free-flow distance.

[0017] In another aspect of the present disclosure, the trigger event is a dynamic traffic light; wherein the communication data includes the current state of the dynamic traffic light and the time interval until the traffic light changes to the next state.

[0018] In another aspect of the present disclosure, the tangible, non-transitory, machine-readable medium further includes machine-readable instructions that, when executed by a processor, cause the processor to input the predicted speed curve back into the ML model to continuously train the ML model.

[0019] In another aspect of the present disclosure, the tangible, non-transitory, machine-readable medium further includes machine-readable instructions that, when executed by a processor, cause the processor to update the personalized driver profile with the predicted speed curve.

[0020] According to certain aspects, a system for driving a host vehicle is provided. The system includes a plurality of vehicle sensors configured to collect external sensor data, vehicle state data, and communication data; a control module configured to analyze the external sensor data, vehicle state data, and communication data to detect a trigger event, a state of a target vehicle, an achievable speed range, and an instantaneous traction force; a decoupling estimator module configured to analyze the trigger event, the state of the target vehicle, and the achievable speed range to determine a free flow distance and an arrival speed at the free flow distance; a pre-trained machine learning (ML) model configured to predict a speed profile of the host vehicle during an approach to a departure event based on the determined free flow distance and the determined arrival speed at the free flow distance; and a cruise control system configured to implement the predicted speed profile.

[0021] In another aspect of the present disclosure, the ML model includes one of a deep neural network, a convolutional deep neural network, a deep belief network, and a recurrent neural network.

[0022] In another aspect of the present disclosure, the system further includes a database configured to store a retryable and updatable personalized driver profile. The decoupling estimator module is further configured to analyze the personalized driver profile to determine the free flow distance and the arrival speed at the free flow distance.

[0023] In another aspect of the present disclosure, the trigger event is a dynamic traffic light. The communication data includes a current state of the dynamic traffic light and a time interval until the traffic light changes to the next state.

[0024] In another aspect of the present disclosure, the decoupling estimator module includes a statistical model configured to perform a statistical analysis based on the trigger event, the state of the target vehicle, and the achievable speed range to determine the free flow distance and the arrival speed at the free flow distance within a predetermined confidence value threshold.

[0025] Further application areas will become apparent from the description provided herein. It should be understood that the description and specific examples are for illustrative purposes only and are not intended to limit the scope of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The drawings described herein are for illustrative purposes only and are not intended to limit the scope of the present disclosure in any way.

[0027] Figure 1 is a functional diagram of a host vehicle according to an exemplary embodiment, the host vehicle having a road driving optimization system for decoupling vehicle state and traffic factors;

[0028] Figure 2 is a plan view of an exemplary traffic scenario according to an exemplary embodiment;

[0029] Figure 3 is a flowchart of a road driving optimization method with decoupling of vehicle state and traffic factors according to an exemplary embodiment; and

[0030] Figure 4 is a hypothetical predicted speed curve of a host vehicle approaching a dynamic traffic signal according to an exemplary embodiment. DETAILED DESCRIPTION

[0031] The following description is merely exemplary in nature and is not intended to limit the present disclosure, application, or uses. The illustrated embodiments are disclosed with reference to the accompanying drawings, wherein like numerals indicate corresponding parts throughout the several views. The drawings are not necessarily to scale, and some features may be enlarged or minimized to show details of particular features. The specific structural and functional details disclosed are not to be construed as limiting, but rather as a representative basis for teaching one skilled in the art how to practice the disclosed concepts.

[0032] As used herein, the terms module, component module, control module, or controller refer to any hardware, software, firmware, electronic control component, processing logic, and / or processor device used alone or in any combination, including but not limited to: application specific integrated circuits (ASICs), electronic circuits, processors (shared, dedicated, or grouped) and memories that execute one or more software or firmware programs, combinational logic circuits, and / or other suitable components that provide the described functionality.

[0033] Embodiments of the present disclosure may be described in terms of functional and / or logical block components and various processing steps. It should be understood that these block components may be implemented by any number of hardware, software, and / or firmware components configured to perform the specified functions. For example, an embodiment of the present disclosure may employ various integrated circuit components, such as memory elements, digital signal processing elements, logic elements, look-up tables, etc., which may perform various functions under the control of one or more microprocessors or other control devices. Additionally, those skilled in the art will appreciate that embodiments of the present disclosure may be practiced in conjunction with any number of systems, and the systems described herein are merely exemplary embodiments of the present disclosure.

[0034] The connecting lines shown in the various figures contained herein are intended to represent example functional relationships and / or physical couplings between the various elements. Conventional techniques may be used for signal processing, data transfer, signaling, control, and other functional aspects of the system (as well as the various operating components of the system), which may not be described in detail herein. It should be noted that there may be many alternative or additional functional relationships or physical connections in the embodiments of the present disclosure.

[0035] The following disclosure provides a road driving optimization system and method for decoupling vehicle states and traffic factors. Driving optimization includes improving fuel economy by minimizing fuel consumption and improving passenger comfort by minimizing jerk, where jerk includes rapid, sharp, and sudden vehicle movements in response to triggering events. Triggering events are traffic events that may require the host vehicle to adjust its speed profile and / or driving path. Non-limiting examples of triggering events include an upcoming dynamic traffic signal, an adjacent target vehicle, and traffic conditions on the host vehicle's path such as a vehicle accident or road closure.

[0036] The influence of the surrounding traffic can be concisely represented by some basic parameters derived from vehicle sensor data, so as to train a machine learning model using a deep learning network and develop a predictive speed profile model for the adaptive cruise control (ACC) system of the host vehicle. In addition to deep learning-based optimization for short-distance performance, incremental corrections can be derived and applied to the optimized speed profile to achieve optimal performance over longer distances. The speed profile can be continuously updated and accumulated for iterative model training, thus generating a more favorable environmental representation, preference estimation, and performance prediction.

[0037] Figure 1 It is a functional diagram of a host vehicle 100 with an intelligent system such as an adaptive cruise control (ACC) system 102, which can be part of an advanced driver assistance system (ADAS) and / or an autonomous driving system (ADS) and is capable of operating between level 0 (no driving automation) and level 5 (full driving automation) according to the SAE J3016 driving automation levels. The vehicle 100 generally includes a body 106, front wheels 108, and rear wheels 110. The body 106 substantially encloses the systems and components of the vehicle 100. The front wheels 108 and the rear wheels 110 are rotatably coupled to the body 106 near the respective corners of the body 106. Although the connected vehicle 100 is shown as a sedan, it is foreseeable that the connected vehicle 100 can be another type of road vehicle, such as a pickup truck, a coupe, a sport utility vehicle (SUV), a recreational vehicle (RV), and a motorcycle.

[0038] As shown in the figure, vehicle 100 generally includes a propulsion system 120, a transmission system 122, a steering system 124, a braking system 126, a detection system 128, a vehicle communication system 130, and various vehicle actuators 132 for operating the components of the vehicle systems 120, 122, 124, 126, 128, 130. The ACC system 102 is configured to cooperate with the vehicle systems 120, 122, 124, 126, 128, 130 and the actuators 132 to control the longitudinal distance between the host vehicle 100 and a target vehicle by controlling the acceleration or deceleration of the host vehicle 100, to keep the host vehicle 100 within a lane by controlling the lateral position of the host vehicle 100, and to control the speed of the host vehicle 100 when the host vehicle 100 approaches a trigger event.

[0039] The host vehicle 100 includes a plurality of sensors 140A-140D, which are configured to collect information and generate sensor data indicative of the collected information. As a non-limiting example, the plurality of sensors 140A-140D includes, but is not limited to, a navigation sensor 140A, which includes a global navigation satellite system (GNSS) transceiver or receiver; a vehicle state sensor 140B, which includes a yaw rate sensor, a speed sensor, and a wheel torque sensor 140B; an external sensor 140C, which includes cameras, lidar, radar, and ultrasonic sensors; and an internal sensor 140D, including an in-vehicle camera. The GNSS transceiver 140A or receiver is configured to detect the position and orientation of the host vehicle 100. The wheel torque sensor 140B is arranged to measure the torque output or torque of the drive wheels 108, 110. The detection range of the external sensor 140C is large enough to detect and identify objects in front of, behind, and to the sides of the host vehicle 100. The internal sensor 140D can detect the alertness or attention state of the vehicle operator and / or passengers.

[0040] The vehicle communication system 130 may include one or more communication transceivers 137, which are configured to wirelessly transmit and receive information or data with other remote entities, such as other networked vehicles using vehicle-to-vehicle (V2V) communication, infrastructure units using vehicle-to-infrastructure (V2I) communication (such as roadside units (RSUs) and mobile edge computing (MEC)), and / or cloud computing service providers 150 using telecommunications. The communication transceiver 137 may be configured to communicate using the IEEE802.11 standard or via a wireless local area network (WLAN) using cellular data communication. However, additional or alternative communication methods, such as dedicated short range communication (DSRC) channels, are also considered within the scope of the present disclosure. A DSRC channel refers to a one-way or two-way short-range to medium-range wireless communication channel designed for automotive use, as well as a corresponding set of protocols and standards.

[0041] The ACC system 102 includes a control module 134 that communicates with one or more vehicle systems 120, 122, 124, 126, 128, 130, vehicle sensors 140A - 140D, and vehicle actuators 132 using Controller Area Network (CAN) and / or Ethernet. The control module 134 is configured to collect navigation data, primary vehicle status data, external environment data, operator / passenger data, and / or traffic data collected by the plurality of sensors 140A - 140D. Non - limiting examples of traffic data include external weather, road conditions, traffic congestion, and dynamic traffic signal information from other connected vehicles, RSUs, cloud - based sources, and / or other sources. The collected data is processed to identify trigger events and predict an optimized speed profile of the primary vehicle 100 when approaching a trigger event. The predicted optimized speed profile improves fuel economy while minimizing the jerkiness of the primary vehicle 100. The control module 134 includes at least one processor 144 and a non - transitory computer - readable storage device or medium 146. The non - transitory computer - readable storage device or medium 146 includes machine - readable instructions that, when executed by the processor 144, cause the processor 144 to perform the method 300 described below.

[0042] The system processor 344 can be a custom or commercially available processor, a central processing unit (CPU), a graphics processing unit (GPU), an auxiliary processor among several processors associated with the control module 134, a semiconductor - based microprocessor (in the form of a microchip or chipset), a macroprocessor, a combination thereof, or generally, a device for executing instructions. For example, the vehicle computer - readable storage device or medium 146 can include volatile and non - volatile memories such as read - only memory (ROM), random access memory (RAM), and keep - alive memory (KAM). KAM is a persistent or non - volatile memory that can be used to store various operating variables when the processor 144 is powered off. The vehicle computer - readable storage device or medium 146 of the controller module 134 can be implemented using a variety of storage devices, such as PROM (programmable read - only memory), EPROM (electrical PROM), EEPROM (electrically erasable PROM), flash memory, or other electrical storage devices, magnetic storage devices, optical storage devices, or combination storage devices that can store data, some of which represent executable instructions used by the vehicle controller 243 in controlling the primary vehicle 200.

[0043] Figure 2 is an illustration of a non - limiting example of a traffic scenario with multiple trigger events. Figure 2A plan view of a road intersection 200 with a dynamic traffic signal 202 is shown. The dynamic traffic signal 202 is configured to manage the vehicle traffic flow through the road intersection 200. The road intersection 200 is defined by a first road 204 intersecting a second road 206. Both the first road 204 and the second road 206 are configured for two-way vehicle travel. The first road 204 includes a road marking 208, as shown by a white solid line 208, which is perpendicular to the first road 204 and is located before the intersection 200 in the direction of travel towards the intersection 200. The white solid line 208 is referred to as a stop line 208 and designates a stop position, which is the stop position for a vehicle in response to a stop signal or command issued by the traffic signal 202.

[0044] The dynamic traffic signal 202 is disposed at the road intersection 200 and is visible to vehicles approaching the intersection 200 on the first road 204 and the second road 206. The traffic signal 202 is capable of sequencing between visual indications (also referred to as phases or states) to manage the vehicle traffic flow through the intersection 200. Common visual indications can include text, symbols, and / or colors. In a non-limiting example, the color indications include a green state, a yellow state, and a red state to respectively indicate that a vehicle may continue through the intersection 200, prepare to stop before the intersection 200, or stop at the intersection 200.

[0045] At least one roadside unit (RSU) 226 is disposed near the intersection 200. The RSU 226 is configured to collect road condition data leading to the traffic lights and transmit the data to the host vehicle 100 and / or a traffic management center. The RSU 226 may be configured to communicate with a remote server, such as a server located in the cloud or back-end, and / or a cellular infrastructure to upload or obtain data related to the safe operation of the intersection. The RSU 226, the remote server, and / or the cellular infrastructure may wirelessly transmit intersection data to connected vehicles operating within a predetermined boundary around the intersection 200. These intersection data may include, but are not limited to, the location of the intersection, the status of the traffic lights, the time to the next state of the traffic lights, information about vehicles near the intersection 200, road conditions, weather conditions, sun glare, etc. The RSU 226 may be configured to communicate with connected vehicles approaching the intersection using vehicle-to-infrastructure (V2I) communication, wireless telematics services, and / or the Internet. Although an RSU is used as an example, a mobile edge computer (MEC) or other infrastructure unit configured to relay such data may also be utilized.

[0046] The figure shows the host vehicle 100 approaching an intersection on the first road 204. The first target vehicle 224 is shown as stopped at the solid line before the intersection, and the second target vehicle 225 is shown as following closely behind the host vehicle 100. The traffic light 202 is shown in the red state (stop command). The traffic light 202 can wirelessly transmit information to the host vehicle 100 that indicates the time period until the traffic light 202 changes from the red state to the green state (go command). In one case, the host vehicle 100 may need to change its speed profile by decelerating to a stop before reaching the solid line 208. In another case, the host vehicle 100 can time the traffic light 202 by adjusting its speed profile when approaching the traffic light 202 so as to drive through the intersection 200 when the traffic light changes to the green state. In both cases, the host vehicle 100 preferably performs the change in speed profile so as to minimize fuel consumption and minimize vehicle jerk, including rapid, sharp, and sudden vehicle movements.

[0047] The target vehicle 224 is shown on the path of the host vehicle 100 between the traffic light and the host vehicle 100. The target vehicle 224 can be considered a higher-priority trigger event that requires changing the profile of the host vehicle 100 before reaching the traffic light. In this case, the host vehicle 100 will execute the following method 300 to determine whether the traffic light 202 or the target vehicle 224 will be considered the primary trigger event based on the lesser of a first free distance between the host vehicle 100 and the target vehicle 224 and a second free distance between the host vehicle 100 and the traffic light 202. The ACC system is configured to respond to higher-priority trigger events and monitor other potential trigger events.

[0048] Figure 3 is a flowchart of a road driving optimization method 300 with decoupled vehicle state and traffic factors according to an exemplary embodiment. The method 300 performs iterative host vehicle 100 speed planning for a short distance ahead based on an estimate of the maximum free flow distance and the corresponding estimated arrival speed at that distance. The maximum free flow distance is estimated with sufficient confidence (e.g., greater than or equal to 0.8) at a constant arrival speed based on one or more influencing trigger events such as traffic light signal phase, surrounding traffic, and host vehicle sensor capabilities. The main components of the method 300 include: (1) decoupling the surrounding traffic factors into the maximum free flow distance and the corresponding arrival speed; (2) selecting a suitable deep learning model based on rules; (3) deriving an optimized short-distance speed profile based on the selected machine learning model; and (4) training the machine model using iteratively updated historical data.

[0049] Figure 4An example of a hypothetical speed curve 400 of the host vehicle 100 approaching a triggering event such as a dynamic traffic light is shown. The horizontal X-axis represents the distance (D1) of the host vehicle 100 to the traffic light 202. The vertical Y-axis represents the speed of the host vehicle 100. In one case, the dashed line 402 represents the change in its speed and speed curve when the host vehicle 100 stops at D1, assuming the traffic light is red (i.e., stopped) when the host vehicle 100 arrives. In another case, the host vehicle 100 decelerates and then resumes speed, and thus it is expected that the traffic signal 202 will change to green (i.e., a go signal) once the host vehicle 100 reaches D1. The solid line 404 represents the speed change of the host vehicle 100 in this case.

[0050] Method 300 begins at block 302 where the host vehicle 100 is traveling on a road. At block 302, the external sensor 140C collects external sensor data about the external environment of the host vehicle 100, the vehicle state sensor 140D collects host vehicle state data including the torque output of the drive wheels and vehicle motion (e.g., driving direction, yaw, pitch, etc.), and the communication system 130 collects communication data including information about weather, road conditions, and traffic conditions from infrastructure units capable of V2I communication and other vehicles capable of V2V communication. The external sensor data, host vehicle state data, and communication data are collectively referred to as collected data or acquisition data.

[0051] At block 304, the control module 134 analyzes the collected data to detect potential triggering events in the travel path of the host vehicle 100. In a non-limiting example, the potential triggering event is a dynamic traffic light in the travel path of the host vehicle 100. The current state (e.g., red, yellow, or green state) of the dynamic traffic light and the timing before changing to the next state can be collected from an infrastructure unit (such as directly from the dynamic traffic light or from an RSU).

[0052] At block 306, the collected information is analyzed to determine the state of target vehicles near the host vehicle 100, including the position of each target vehicle relative to the host vehicle 100, the driving direction of each target vehicle, and the speed of each vehicle. When estimating the free flow distance and the arrival speed at the free flow distance, the target vehicles immediately in front of and behind the host vehicle 100 are considered. The estimated distance and arrival speed of the host vehicle 100 follow the condition that the host vehicle 100 maintains a safe distance from the front target vehicle 224 and with a high enough probability avoids sudden deceleration of the rear target vehicle 225 due to sudden changes in speed.

[0053] At block 308, the collected information is analyzed to determine a feasible speed range based on the configuration of the host vehicle, the segment speed limit of the road, road conditions, weather conditions, etc.

[0054] At block 310, the collected information is analyzed to determine the instantaneous traction force. The instantaneous traction force can be determined based on the road condition information obtained by the torque sensor 140B, the external sensor 140C, and the primary vehicle motion information collected by the vehicle state sensors.

[0055] At block 303, a personalized operator profile is retrieved from the database.

[0056] At block 312, the determined trigger event, the determined state of one or more target vehicles, the determined feasible speed range, and the retrieved personalized operator profile are input into the decoupled estimator module 148. The decoupled estimator module 148 is configured to incorporate or reduce a complex scenario including the determined trigger event, the determined target vehicle state, and the feasible speed range into essentially two representative variables, including (i) the maximum free flow distance at block 314 and (ii) the arrival speed at the free flow distance at block 316.

[0057] The decoupled estimator module 148 includes a statistical model established for the behavior of surrounding target vehicles, obtains state data that may affect the surrounding target vehicles, and performs a statistical analysis to evaluate (i) the maximum free flow distance and (ii) the arrival speed at the free flow distance with a predetermined confidence value threshold. The decoupled estimator module 148 is configured to monitor relatively well-defined trigger events, such as traffic light timing and lane closures ahead, so as to incorporate relevant constraints on the maneuvering of the primary vehicle 100.

[0058] Continuing with the traffic light example, the decoupled estimator module 148 estimates the maximum distance from the current position of the primary vehicle 100 to the stop line 208 at the road intersection 200 where the stop light 202 is located or to the target vehicle 224 immediately in front of the primary vehicle 100. The arrival speed can be the estimated speed for a complete stop in response to the traffic light 202 indicating a red signal or the estimated speed for passing through the intersection 200 in response to the traffic light 202 being green or yellow when the primary vehicle reaches the intersection 200.

[0059] Proceed to block 318, where the control module 134 determines whether the maximum free flow distance is greater than a predetermined free flow distance threshold. If the determined free flow distance is less than the predetermined free flow distance threshold, the method 300 ends and the primary vehicle 100 maintains its current ACC system, derived to maintain a safe distance from the target vehicle 224 ahead or to the trigger event. If the determined maximum free flow distance is greater than the predetermined free flow distance threshold, the method 300 proceeds to block 320.

[0060] At block 320, the determined (i) maximum free flow distance, (ii) arrival speed at the free flow distance, and (iii) instantaneous traction force are fed into a machine learning (ML) model to plan a speed profile subject to feasibility limits of speed and acceleration at a short distance in front of the host vehicle 100. The term short distance is defined as the distance in the travel path of the host vehicle 100 with a controllable level of uncertainty. Given the way the ML model is trained, the optimal short distance target speed can be performed by evaluating a set of discrete speed values within the allowed speed range based on a customized metric of the ML model and selecting the speed value corresponding to the best performance. Measuring the short distance by distance is compatible with the comparison of most customized metrics. The ML model can be selected to reflect driving behaviors (smooth, slow moving, congested, etc.) and surrounding traffic conditions in a range of situations, and the surrounding traffic conditions are classified into several types according to the surrounding traffic parameters estimated in blocks 310 and 314.

[0061] Machine learning includes a training phase and an inference phase. In the training phase, when input data is input into the machine learning model, the machine learning model adjusts its weights until an appropriate fit is obtained. In the inference phase, the trained machine learning model can make inferences based on real-time data. The inference model predicts a speed profile when the host vehicle 100 approaches a trigger event based on the trained machine learning model. The training phase can be performed on a server outside the vehicle, such as a cloud server or a server at a remote location (also known as the backend). When the host vehicle 100 approaches a trigger event, the speed profile generated by the inference model is executed by the ACC system to control the host vehicle 100.

[0062] When possible, the optimization of the speed profile of the host vehicle 100 may involve the training of one or more deep learning models in combination with rule-based selection of optimization methods and the derivation and application of incremental corrections to the optimization results to favor long-distance performance, with the goal of deriving an optimal speed profile within a short distance ahead to optimize certain performance metrics (e.g., minimizing energy consumption) within that range. The ML model is trained based on historical data such as the saved maximum free flow distance, arrival speed at that distance, actual speed achieved at short distance, instantaneous traction force, actual energy consumption at short distance, etc.

[0063] At block 322, the ML model can include a deep neural network, a convolutional deep neural network, a deep belief network, and / or a recurrent neural network, which are configured to evaluate the selected one or more performance metrics for various speed values in a given situation.

[0064] At block 324, the ML model outputs a short-distance optimized speed curve by combining deep learning with rule-based model selection for certain local performance metrics such as energy efficiency and maximum jerk, and then applying the derived incremental correction helps to achieve optimal performance from a longer-distance perspective so that some satisfactory trade-offs can be achieved between complexity and performance in the case of uncertain traffic factors.

[0065] Proceed to block 326, and transmit the short-distance optimized speed curve to the ACC system. When the host vehicle 100 approaches a trigger event, the ACC system executes the speed curve.

[0066] Proceed to block 328, and update the historical speed curve with the newly developed speed curve. In addition to the deep learning-based optimization for short-distance performance, incremental corrections can also be derived and applied to the optimized speed curve to achieve optimal performance at longer distances. During the use of this scheme, driving data will be continuously updated and accumulated for iterative model training at blocks 320 and 322, so as to obtain a more favorable environmental characterization, preference estimation, and performance prediction of the speed curve.

[0067] Starting from block 328, save the developed speed curve as part of the personalized curve in block 303. The method returns to block 302 and continues to monitor the maximum free-flow distance and the corresponding arrival speed.

[0068] The trigger event can be interpreted as a bridge between human (or system) perception and algorithmic methods, which helps to more intuitively understand the decoupled processing part of method 300. In the case of manual driving, the driver continuously observes the surrounding objects (traffic lights, pedestrians, other vehicles, etc.) to notice the occurrence of any trigger event. Although the driver perceives based on the occurrence of the event (the traffic light turns red, a vehicle cuts in, etc.), the actual situation that determines the need to take corresponding actions is the change in the surrounding traffic variables caused by the event. Therefore, the variables "free-flow distance" and "arrival speed" summarize all the influences related to the host vehicle 100 and can be applied to all types of trigger events, which are intended to represent the driver's or system's reaction to the surrounding trigger events through the input of the ML model.

[0069] The above-disclosed systems and methods provide a cost-effective way to derive the optimal vehicle speed curve in the upcoming short distance and achieve a satisfactory trade-off between complexity and performance in the case of uncertain traffic factors. The description of the present disclosure is merely exemplary in nature, and variations that do not depart from the gist of the present disclosure are intended to fall within the scope of the present disclosure. These variations should not be regarded as departing from the spirit and scope of the present disclosure.

Claims

1. A method of driving a host vehicle, comprising: Receiving, by a control module, data collected by at least one vehicle sensor; Analyzing, by the control module, the collected data to detect a trigger event and determine a feasible speed range; Determining, by a decoupled estimator module, a free flow distance and an arrival speed at the free flow distance based on the detected trigger event and the determined feasible speed range; Inputting the determined free flow distance and the determined arrival speed at the free flow distance into a pre-trained machine learning (ML) model to predict a speed curve to the trigger event; And Steering the host vehicle towards the trigger event based on the predicted speed curve.

2. The method according to claim 1, further comprising: Analyzing the collected data to determine a state of a target vehicle, wherein the determined state of the target vehicle includes a position of the target vehicle relative to the host vehicle and a speed of the target vehicle; Wherein the decoupled estimator module further determines the free flow distance and the arrival speed at the free flow distance based on the determined state of the target vehicle.

3. The method according to claim 2, further comprising inputting an incremental correction of the determined free flow distance and the determined arrival speed at the free flow distance into the pre-trained ML model, thereby iteratively predicting the speed curve.

4. The method according to claim 1, further comprising: Obtaining a personalized operator profile; Wherein the decoupled estimator module further determines the free flow distance and the arrival speed at the free flow distance based on the obtained personalized operator profile.

5. The method according to claim 1, further comprising: Determining an instantaneous traction force of the host vehicle; And Inputting the determined instantaneous traction force together with the determined free flow distance and the determined arrival speed at the free flow distance into the pre-trained ML model to predict the speed curve to the trigger event.

6. The method according to claim 1, wherein The predicted speed curve minimizes fuel consumption and minimizes jerk motion of the host vehicle during the process of reaching the trigger event.

7. The method according to claim 1, wherein Steering the host vehicle towards the trigger event by an adaptive cruise control (ACC) system based on the predicted speed curve.

8. The method according to claim 1, wherein, The determined free flow distance is greater than a predetermined free flow distance threshold.

9. The method according to claim 1, wherein Training the pre-trained machine learning (ML) model on a deep neural network.

10. The method according to claim 1, wherein, The trigger event is a dynamic traffic light, wherein the collected data includes a current state of the dynamic traffic light and a time interval until the dynamic traffic light changes to a next state.

Images