DEVICES, SYSTEMS AND METHODS FOR PLANNING INTERSECTION TURNS

Abstract

The presented devices, systems, and methods are directed toward trajectory planning. Example devices include one or more processors operable to receive an instruction to turn into an intersection, plan a trajectory for an autonomous vehicle to turn into the intersection based on map data including one or more optimal turn paths associated with the intersection, and instruct the autonomous vehicle to follow the trajectory to negotiate the intersection. The one or more optimal turn paths are generated based on turn paths of one or more human-driven vehicles.

Description

TECHNICAL FIELDThe present disclosure relates to apparatuses, systems, and methods for controlling vehicles, and more particularly, to apparatuses, systems, and methods for controlling vehicles by planning intersection turns.BACKGROUNDIntersections present a challenge to vehicles because they fight around the same space. At intersections, there are several lanes, oncoming traffic and traffic signs at which the forward lane must be determined. Relying on real-time decisions at intersections can result in uncertainties. Therefore, planned intersection turns for vehicles must be established to reduce potential traffic risks and increase efficiency in driving intersections.SUMMARYIn one embodiment, a trajectory planning apparatus includes one or more processors operable to receive an instruction to turn within an intersection, plan a trajectory for an autonomous vehicle to turn within the intersection based on map data having one or more optimal turn paths associated with the intersection, wherein the one or more optimal turn paths are generated based on turn paths from one or more human-driven vehicles, and direct the autonomous vehicle to follow the trajectory to pass through the intersection.In another embodiment, a method of trajectory planning includes receiving an instruction to turn within an intersection, planning an autonomous vehicle trajectory to turn within the intersection based on map data having one or more optimal turns associated with the intersection, the one or more optimal turns generated based on turns of one or more human-driven vehicles, and instructing the autonomous vehicle to follow the trajectory to pass through the intersection.These and other features of the embodiments of the present disclosure will be better understood in view of the following detailed description taken in conjunction with the drawings.BRIEF DESCRIPTION OF THE DRAWINGSThe embodiments illustrated in the drawings are illustrative and exemplary and are not intended to limit the disclosure. The following detailed description of the embodiments will be understood when read in conjunction with the following drawings, in which like structures are represented by like reference numerals, and in which: FIG. 1 schematically illustrates an example of a system for trajectory planning of intersection turns according to one or more embodiments shown and described herein; FIG. 2 schematically illustrates example components of the apparatus and system for trajectory planning of intersection turns of the present disclosure, in accordance with one or more embodiments shown and described herein; FIG. 3 illustrates an illustrative block diagram for generating the planned trajectory at an intersection according to one or more embodiments illustrated and described herein; and FIG. 4 illustrates a flow chart illustrating the steps for generating the planned intersection turn operations at the intersection according to one or more embodiments shown and described herein.DETAILED DESCRIPTIONThe disclosed embodiments include apparatuses, systems, and methods for trajectory planning at intersections based on map data including one or more optimal turn paths. The optimal turn paths may be generated based on the turn paths of one or more human-driven vehicles. The disclosed embodiments include apparatuses, systems, and methods useful for autonomous vehicles in planning trajectories for passing intersections. Autonomous vehicles that rely on real-time decisions at intersections may have difficulty navigating through an intersection, resulting in undesirable wide turns where the autonomous vehicles are going out to the intersection. Such wide turns may interfere with traffic flow and increase the risk of collision with other vehicles, pedestrians, or cyclists within the intersection.By encoding the optimal turn path of human-driven vehicles into sections within map data, the autonomous vehicles can turn more naturally and desirably within the intersection. Disclosed embodiments further include apparatuses, systems, and methods having artificial intelligence functions to continuously improve the optimal turn paths based on the generated scheduled trajectory. Accordingly, the disclosed apparatuses, systems, and methods are useful for mitigating issues associated with navigation of vehicles passing an intersection, such as wide turns that may reach the intersection, and thereby improve overall traffic flow and interaction of the vehicles. Moreover, the artificial intelligence functions improve the system's capabilities by continuously refining optimal turn paths based on generated scheduled trajectories to ensure that the trajectory planning system continues to develop, account for traffic dynamics changes, and contribute to efficient and desirable safe navigation of intersections by autonomous vehicles, and ensure that the trajectory planning remains robust and responds to developing road conditions, traffic patterns, and other dynamic factors. The trajectory planning system not only gives foreground safety and natural maneuvering, but also demonstrates efficiency in using map data. The integration of optimal turn paths into map data facilitates accurate decision making for autonomous vehicles and uses spatial information to effectively navigate intersections.Various embodiments of the methods and systems for trajectory planning are described in more detail here. Wherever possible, the same reference numbers will be used in the drawings to refer to the same or like parts. As used herein, the singular forms "a", "an" and "the" include plural referents unless the context clearly dictates otherwise. For example, reference to "a" component includes aspects having two or more such components, unless the context clearly indicates otherwise.FIGS. 1 and 2 schematically show an example of a system for trajectory planning 100. The trajectory planning system 100 may include one or more controllers 201, which may also include one or more modules, such as a trajectory generation module 222 and an optimal turn path module 232.The trajectory generation module 222 may include one or more first machine learning algorithms, such as first neural networks 322. The trajectory generation module 222 may generate a planned trajectory 105. The planned trajectory 105 may include an upcoming turn path, as well as speed, time, and kinematics of the vehicle 101 associated with the upcoming turn path.The optimal turn path module 232 may include one or more second machine learning algorithms, such as second neural networks 432. The second machine learning algorithm may generate one or more optimal turn paths 315 (as shown in FIG. 3 ) associated with an intersection 135. The information about the optimal turn paths 315 and the intersection 135 may be stored in map data 217. The map data 217 may include pedestrian crossings, traffic lights, traffic signs, obstacles, lanes, road edges, shoulder, dividers, color markers, posts, or a combination thereof. In embodiments, the map data 217 may be high resolution map data (HD) or standard resolution map data (SD). An SD map including SD map data is based on curves, elevations, and coordinates of roads. An SD card may have a resolution of one meter. An HD map with HD map data may contain more details than an SD map with SD map data. The HD map may have a resolution of less than one meter to one centimeter. The HD map data may include all information also included in the SD map data, e.g., road shape, road markings, traffic signs, poles, guardrails, walls, and obstacles.The trajectory planning system 100 may include one or more vehicles 101, which may be autonomous vehicles. In some embodiments, the one or more controllers 201 are included in the vehicles 101. In some embodiments, some of the vehicles 101 may include communication devices, such as vehicle network interface hardware, operable to wirelessly communicate with the controllers 201. In some embodiments, the controller 201 may be included in one or more servers that include server communication devices, such as network interface hardware 206, operable to communicate with the vehicles 101.Each of the vehicles 101 may be an automobile or other vehicle with or without passengers, such as a land vehicle, water, and / or air vehicle. Each of the vehicles 101 may be an autonomous vehicle that navigates its environment with limited human input or without human input. Each of the vehicles 101 may travel on a road and perform vision-based lane centering, e.g., using one or more sensors. Each of the vehicles 101 may include actuators for propelling the vehicle, e.g., an engine, an engine, or another powertrain. The vehicles 101 may travel on various surfaces, such as without limitation on roads, freeways, expressways, bridges, tunnels, parking spaces, garages, terrain routes, rails, or other surfaces on which the vehicles may operate.In embodiments, the vehicles 101 may move on a road 137 that includes one or more intersections 135. The intersections 135 may include one or more lanes. The intersections 135 may include traffic signs, signals, circumcircle, and other structures for controlling traffic flow. The intersections 135 may be four-lane, intersection-free, three-lane (e.g., T-intersection and Y-intersection), or five- or more-lane. As illustrated in FIG. 1, the road 137 may include, for example, a vertical road and a horizontal road. The vertical road may include two lanes, namely a lane 121 toward the south and a lane 122 toward the north. The horizontal road may include two lanes, a west lane 131 and an east lane 132. The vertical road and the horizontal road may cross at the intersection 135. The intersection 135 may include one or more curbs 125 at the joints of the vertical road and the horizontal road.FIG. 2 schematically illustrates example components of the controller 201. Although FIG. 2 illustrates a controller 201, in some embodiments, the trajectory planning system 100 may also include two or more controllers 201. The controller 201 or vehicles 101 may include one or more vision sensors 208 and vehicle sensors 212. Vision sensors 208 may be used to capture images or videos of the environment of vehicles 101. In some embodiments, the one or more vision sensors 208 include one or more imaging sensors configured to operate in the visual and / or infrared spectrum to detect visual and / or infrared light. Although the embodiments described herein relate to hardware for detecting light in the visual and / or infrared spectrum, other types of sensors are also conceivable. For example, the systems described herein could include one or more LIDAR sensors, radar sensors, sonar sensors, or other types of sensors for collecting data that can be integrated into or supplement the data collection described herein. Range sensors such as radar may be used to obtain rough depth and speed information for the vision of the vehicle 101. The one or more sensors 208 may include a forward facing camera installed in the vehicles 101. The one or more sensors 208 may be any device having an array of sensors capable of sensing radiation in an ultraviolet wavelength range, a visible light wavelength range, or an infrared wavelength range. The one or more sensors 208 may have any resolution. In some embodiments, one or more optical components, such as a mirror, a fisheye lens, or other type of lens, may be optically coupled to the one or more vision sensors 208. In the embodiments described herein, the one or more vision sensors 208 may provide image data to the one or more processors 204 or another component communicatively connected to the communication path 203. In some embodiments, the one or more vision sensors 208 may also provide navigation assistance. That is, data captured by the one or more vision sensors 208 may be used for autonomous or semi-autonomous navigation of a vehicle.The controller 201 or the vehicles 101 may include one or more vehicle sensors 212. Each of the one or more vehicle sensors 212 is coupled to the communication path 203 and communicatively connected to the one or more processors 204. The one or more vehicle sensors 212 may include one or more speed sensors or motion sensors for detecting and measuring motion and changes in motion of a vehicle, e.g., the vehicle 101. The motion sensors may comprise inertial measurement units. Each of the one or more motion sensors may include one or more accelerometers and one or more gyroscopes. Each of the one or more motion sensors converts the sensed physical motion of the vehicle into a signal indicative of an orientation, a rotation, a speed, or an acceleration of the vehicle. The data captured by the vehicle sensors 212 may be used to determine the kinematics of the vehicle 101.The image sensors 208 and the vehicle sensors 212 can be used to capture data for controlling the vehicle, data relating to the road state and data relating to the kinematics of the vehicle. The vehicle control data, the road condition data, and the vehicle kinematics data may be used to monitor the actual trajectory of the vehicle 101 as it passes through the intersection 135. The data for controlling the vehicle may include the coasting position, the braking status, the steering angle, and the gear selection of the vehicle 101. The road condition data may include road type, coefficient of friction, and surface irregularities (e.g., bumps). The vehicle kinematics data may include speed, acceleration, position, and orientation of the vehicle 101.The controller 201 may include one or more processors 204. Each of the one or more processors 204 may be any device capable of executing machine readable and executable instructions. The instructions may be in the form of a machine readable instruction set stored in the data storage component 207 and / or a storage component 202. Accordingly, each of the one or more processors 204 may be a controller, an integrated circuit, a microchip, a computer, or another computing device. The one or more processors 204 are coupled to a communication path 203 that establishes a signal connection between the various modules of the system. Accordingly, communication path 203 may communicatively couple any number of processors 204 to each other and allow modules coupled to communication path 203 to operate in a distributed computing environment. In particular, each of the modules may operate as a node that may transmit and / or receive data. As used herein, the term "communicatively coupled" means that coupled components are capable of exchanging data signals with each other, such as electrical signals over a conductive medium, electromagnetic signals over air, optical signals over optical fibers, and the like.Accordingly, the communication path 203 may be formed of any medium capable of transmitting a signal, such as conductive wires, conductive traces, optical fibers, or the like. In some embodiments, communication path 203 may enable transmission of wireless signals, such as WiFi, Bluetooth® near field communication (NFC), and the like. Moreover, the communication path 203 may be formed of a combination of media capable of transmitting signals. In one embodiment, communication path 203 includes a combination of conductive traces, conductive wires, connectors, and buses that cooperate to facilitate the transmission of electrical data signals to components such as processors, memories, sensors, input devices, output devices, and communication devices. Accordingly, the communication path 203 may include a vehicle bus, such as a LIN bus, a CAN bus, a VAN bus, and the like. Moreover, it should be noted that the term "signal" refers to a waveform (e.g., electrical, optical, magnetic, mechanical, or electromagnetic) such as direct current, alternating current, sine wave, triangular wave, square wave, vibration, and the like, which may move through a medium.The controller 201 may include one or more storage components 202 connected to the communication path 203. The one or more memory components 202 may include RAM, ROM, flash memory, hard drives, or any other device capable of storing machine readable and executable instructions such that the machine readable and executable instructions may be retrieved from the one or more processors 204. The machine readable and executable instructions may include logic or algorithms written in any generation programming language (e.g., 1GL, 2GL, 3GL, 4GL, or 5GL), such as machine language that may be executed directly by the processor, or assembly language, object oriented programming (OOP), scripting languages, microcode, etc., that may be compiled or compiled into machine readable and executable instructions and stored on the one or more memory components 202. Alternatively, the machine readable and executable instructions may be written in a hardware description language (HDL), such as logic implemented via either a field programmable gate array (FPGA) configuration or an application specific integrated circuit (ASIC), or their equivalents. Accordingly, the methods described herein may be implemented in any conventional computer programming language, as preprogrammed hardware elements, or as a combination of hardware and software components.The one or more storage components 202 may include one or more modules, including the trajectory generation module 222 and the optimal turn path module 232. Each of the one or more modules may include, but are not limited to, routines, subroutines, programs, objects, components, data structures, and the like for performing particular tasks or for executing particular types of data, as described below. The data storage component 207 may store map data including the intersections 135, the road 137, and the optimal turn paths 315. The data storage component 207 may further store training data 227 for training the trajectory generation module 222 and the optimal turn path module 232. The training data 227 may include ground truth data related to trajectory planning at one or more intersections for human-driven vehicles and autonomous vehicles. The data storage component 207 may store historical data 237, such as historical vehicle kinematics data, historical vehicle control data, historical road condition data, historical trajectory data of the vehicle 101, and other historical data related to operation of the vehicle 101, as well as historical sensor data of the human-driven vehicles 301 (e.g., as shown in FIG. 3 ). The trajectory generation module 222 and the optimal turn path module 232 may also be stored in the data storage component 207 during operation or after operation.The one or more modules, including the trajectory generation module 222 and the optimal turn path module 232, may include one or more machine learning algorithms, such as neural networks. The modules may be trained via a neural network as described herein and equipped with machine learning capabilities. By way of example, and not limitation, the neural network may use one or more artificial neural networks (ANNs). In ANNs, the connections between the nodes may form a directed acyclic graph (DAG). ANNs may include node inputs, one or more hidden activation layers, and node outputs, and may be used with activation functions in the one or more hidden activation layers, such as a linear function, a step function, a logistic (sigmoid) function, a tanh function, a rectified linear unit (ReLu function), or combinations thereof. ANNs are trained by applying such activation functions to training datasets to determine an optimized solution from the adjusted weights and biases applied to the nodes within the hidden activation layers to generate one or more outputs as an optimized solution with a minimum error. In machine learning applications, new input (e.g., the generated output(s)) may be provided to the ANN model as training data to further improve accuracy and minimize the error of the ANN model. The one or more ANN models may use one-to-one, one-to-many, many-to-one, and / or many-to-many sequence modelling (e.g., sequence-to-sequence). The one or more ANN models may use a combination of artificial intelligence techniques, such as deep learning, random forest classifiers, feature extraction from audio, images, clustering algorithms, or combinations thereof. In some embodiments, a convolutional neural network (CNN) may be used. For example, a convolutional neural network (CNN) may be used as an ANN, which is a class of deep forward ANNs used for audio analysis of images in the field of machine learning. CNNs may be shift- or space-invariant and use a shared weight and translation architecture. In addition, each of the various modules may include one or more artificial intelligence generative algorithms. The generative artificial intelligence algorithm may include a general adventious network (GAN) that consists of two networks, a generator model and a discriminator model. The generative artificial intelligence algorithm may also operate based on variation autoencoders (VAE) or transformer-based models.The controller 201 may include input / output hardware 205 connected to the communication path 203. The input / output hardware 205 may include a monitor, keyboard, mouse, printer, camera, microphone, speaker, and / or other device for receiving, transmitting, and / or presenting data. The controller 201 may include network interface hardware 206 for communicatively coupling the controller 201 to external resources (e.g., the vehicles 101 or smart devices), the Internet of Things (IoT), and / or a server. Network interface hardware 206 may be communicatively coupled to communication path 203, and may be any device capable of transmitting and / or receiving data over a network. Accordingly, the network interface hardware 206 may include a communication transceiver for transmitting and / or receiving any wired or wireless communication. For example, the network interface hardware 206 may include an antenna, a modem, a LAN port, a WiFi card, a WiMAX card, mobile communication hardware, near field communication hardware, satellite communication hardware, and / or any wired or wireless hardware for communication with other networks and / or devices. In one embodiment, network interface hardware 206 includes hardware configured to operate the Bluetooth® wireless communication protocol. For example, the network interface hardware 206 of the trajectory planning system 100 may receive and / or transmit map data 217, planned trajectories 105, sensory data of the vehicle 101 (such as speed data, acceleration data, steering data, yaw data, wheel slip data, lane departure data, time, weather conditions, vehicle type, vehicle size, minimum and maximum vehicle turn radii) with a server or vehicles 101.Illustrated in FIGS. 1 and 3 are examples of the trajectory planning system 100 that the trajectory generation module 222 uses to generate the planned trajectory 105 for the vehicle 101 (e.g., an autonomous vehicle) approaching the upcoming intersection 135 to pass the intersection 135 based on one or more optimal turn paths 315 included in the map data 217. FIGS. 1 and 3 further show an example of the trajectory planning system 100 that the optimal turn path module 232 uses to generate one or more optimal turn paths 315 associated with the intersection 135 in the map data 217.In embodiments, the vehicle 101 may recognize the imminent intersection 135 on the road 137 and send an instruction to the controller 201 to assist in turning at the intersection 135. Upon receiving the instruction to turn, the controller 201 may generate the planned trajectory 105 based on one or more optimal turn paths 315 stored in the map data 217 for the vehicle 101, for example, for autonomous vehicles, and instruct the vehicle 101 to follow the trajectory to pass the intersection 135. The planned trajectory 105 may include the imminent turn path as well as the speed, time, and kinematics of the vehicle 101 associated with the imminent turn path. Each imminent turn path may consist of an entry point 151 at a boundary of the intersection 135, an existing point 153 at the boundary of the intersection 135, and a geometric path 152 between the entry point 151 and the existing point 153. In some embodiments, the imminent turn path may be one of the one or more optimal turn paths 315.In some embodiments, the vehicles 101 may use the vision sensors 208 and the vehicle sensors 212 to generate and transmit sensory data, such as the vehicle kinematic data, the vehicle control data, and the road condition data, to the controller 201. In exemplary embodiments, the vehicle kinematic data may be a vehicle position, a vehicle speed and a vehicle acceleration of the vehicle 101. The vehicle control data may be vehicle steering, vehicle thrust, and brake inputs of the vehicle 101. The road condition data may include the road surface condition, the road geometry, and the traffic conditions on the road 137. Upon receiving the sensory data and other input data, such as the local weather conditions, the controller 201 may generate the planned trajectory 105 of the vehicle 101 using the trajectory generation module 222In some embodiments, the planned trajectory 105 may be generated based on parameters of the vehicle 101 and parameters of the human guided vehicles 301 associated with one or more optimal turn paths 315. The vehicle 101 parameters and the human guided vehicle 301 parameters may include vehicle length, minimum turn radii, steering system, acceleration and deceleration performance, or other parameters related to the turn performance at the intersection 135. The trajectory planning system 100 may compare the similarity of the parameters between the vehicle 101 and the human-driven vehicle 301 and adjust the planned trajectory 105 inversely proportional to the similarity. In some embodiments, the trajectory planning system 100 may determine whether the similarity is equal to or greater than a similarity threshold. The trajectory planning system 100 may, in generating the planned trajectory 105, select the optimal turn paths 315 associated with the similarity equal to or greater than the threshold similarity and disregard the optimal turn paths 315 associated with the similarity less than the threshold similarity.In some embodiments, the trajectory planning system 100 may collect the sensor data from the image sensors 208 and the vehicle sensors 212 and continue to operate the vehicle 101 to follow the planned trajectory 105 by controlling or adjusting the steering, thrust, and brake inputs of the vehicle 101. The trajectory planning system 100 may monitor a lane 154 of the vehicle 101 as the vehicle 101 passes through the intersection and determine whether the lane 154 deviates from the planned trajectory 105. In response to determining that the lane 154 deviates from the planned trajectory 105 (e.g., as shown in FIG. 1, the lane 154 deviates from the geometric trajectory 152), the trajectory planning system 100 may generate an updated trajectory for the vehicle 101 to pass the intersection 135 based on historical turns 303 and 305 of the human-driven vehicles 301.As shown in FIG. 3, the trajectory planning system 100 may use the optimal turn path module 232 to generate one or more optimal turn paths 315 associated with the intersection 135 based on turn paths 303 and 305 from one or more human-driven vehicles 301. In some embodiments, the optimal turn paths 315 may be an average turn path of the input turn paths 303 and 305 of the human guided vehicles 301. In some embodiments, the optimal turn paths 315 may be a shortest turn path of the input turn paths 303 and 305 of the human guided vehicles 301. In some embodiments, the optimal turn paths 315 may be the most energy efficient turn path of the input turn paths 303 and 305 of the human guided vehicles 301. The one or more optimal turn paths 315 may be further generated based on historical sensor data of the human-driven vehicles 301 over time. The sensor data may include speed data, acceleration data, steering data, yaw data, lane departure data, wheel slip data, time of day data, weather conditions, vehicle type, vehicle size, minimum and maximum turning radius of the vehicle, or a combination thereof.According to embodiments in FIGS. 2 and 3, the trajectory generation module 222 and the optimal turn path module 232 may include one or more neural networks 322 and 432. Each of the neural networks 322 and 432 may include an encoder, one or more hidden layer layers, and a decoder. Neural networks 322 and 432 may feed training data 227 into the encoder during the pre-training process to generate a low-dimensional representation of the target input-output pairs. For example, the low-dimensional representation may include input of sensor data collected with vision sensor 208 and vehicle sensors 212 coupled to planned trajectories as well as detected trajectories of vehicles 101. The first neural network 322 may output the planned trajectory 105. The second neural network 432 may output an optimal turn path 315 that is integrated into the map data 217.In some embodiments, the trajectory generation module 222 and the optimal turn path module 232 may include one or more neural networks 322 and 432 trained with the training data 227 and the historical data 237. Neural networks 322 and 432 may include the encoder and / or decoder in conjunction with a layer normalization operation and / or an activation function operation. The encoded input data may be normalized and weighted by the activation function before being provided to the hidden layers. The hidden layers may generate a representation of the input data in a bottleneck layer. After the neural network processed data is passed to the last layer of the neural network, global layer normalization may be performed to normalize the scheduled trajectories and optimal turn paths. The outputs may be normalized and converted using an activation function for training and verification purposes, as described in detail below. The activation function may be linear or nonlinear. The activation function may be, without limitation, a sigmoid function, a softmax function, a hyperbolic tangent function (Tanh), or a rectified linear unit (ReLU). Neural networks 322 and 432 may feed the encoder with historical data 237 such as historical vehicle kinematics data, historical vehicle control data, historical road condition data, historical planned trajectory, historical detected trajectory, historical turn paths of man-propelled vehicles, and other historical vehicle operation data for further training.In embodiments, the one or more vehicle modules may be pre-trained using training data 227, including basic examples and scenarios in which multiple entities (e.g., vehicles 101 and human-driven vehicles 301) are traveling on road 137, including multiple lanes 121, 122, 131, and 132, intersections 135, and curbs 125. Pretraining may include identifying the entities and the desired planned trajectories and optimal turn paths based on the entities and intersections 135 in the examples and scenarios, and using one or more neural networks 322 and 432 to learn to predict the desired and undesired trajectories and optimal turn paths based on the training data 227. Pretraining may also include fine tuning, scoring, and testing steps. The modules may be continuously trained using the data collected under real conditions as historical data 237 to adapt to changed conditions and factors and improve performance over time. The neural network may be trained based on activation functions mentioned above. The encoder may generate encoded input data h=(Wx+b) that is transformed from the input data of one or more input channels. The encoded input data of one of the input channels may be represented as h ij= g(Wx ij+ b) from the raw input data x ij which is then used to reconstruct the output x ξ ij= f(W T h ij+ b'). The neural networks may reconstruct outputs such as planned trajectories 105 and optimal turn path 315 in x'=(W T h+b'), where W is the weight, b is the bias voltage, W T and b' are the lateral values of W and b and are learned by backpropagation. In this process, the neural networks for each input file may calculate a distance between an input file x and a reconstructed input file x' to obtain a distance vector |x-x'|. The neural networks 322 and 432 may minimize the loss function, which is a utility function as the sum of all distance vectors. The training process may enable neural networks 322 and 432 to learn linear or non-linear representations of the input data. The accuracy of the predicted output may be evaluated by satisfying a preset value, such as a preset value for the accuracy and the area under the curve (AUC) calculated using an output evaluation of the activation function (e.g., the softmax function or the sigmoid function). For example, the trajectory planning system 100 may rank the preset value of the AUC to be 0.7 to 0.8 as an acceptable simulation, 0.8 to 0.9 as an excellent simulation, or more than 0.9 as an excellent simulation. After training satisfies the preset value, the updated neural networks 322 and 432 may be stored in the trajectory generation module 222 and the optimal turn path module 232, respectively, that are used to generate future planned trajectories and optimal turn paths.FIG. 4 shows a flow diagram of a method 400 for trajectory planning. At block 401, the present method 400 receives an instruction to turn at the intersection 135. At block 403, the present method 400 includes planning a trajectory 105 for the vehicle 101 to turn within the intersection 135 based on one or more optimal turn paths 315 associated with the intersection 135 in the map data 217. The one or more optimal turn paths 315 are generated based on turn paths 303 and 305 from one or more human-driven vehicles 301.In some embodiments, the map data 217 may include pedestrian crossings, traffic lights, traffic signs, shoulder tickets, lanes, road edges, dividers, color markers, posts, or a combination thereof. The map data 217 may be HD map data or SD map data.In some embodiments, the one or more optimal turn paths 315 may be further generated based on historical sensor data of the human-driven vehicles 301 over time. The sensor data may include speed data, acceleration data, steering data, yaw data, wheel slip data, lane departure data, time, weather conditions, vehicle type, vehicle size, minimum and maximum vehicle turn radii, or a combination thereof. The planned trajectory 105 may include an upcoming turn path, as well as speed, time, and kinematics of the vehicle 101 associated with the upcoming turn path.In some embodiments, the planned trajectory 105 may be generated based on a trained first machine learning algorithm (such as the first neural network 322 in FIGS. 2 and 3 ) that compares one or more optimal turn paths 315 and parameters of the vehicle 101 with parameters of the human-driven vehicles 301 associated with the one or more optimal turn paths 315. The parameters of the vehicle 101 and the parameters of the human guided vehicles 301 may include vehicle length, minimum turn radii, steering system, acceleration and deceleration performance, or a combination thereof.In some embodiments, the one or more optimal turn paths 315 may be generated by a second trained machine learning algorithm (such as the second neural network 432 in FIGS. 2 and 3 ) configured to reduce the path lengths of the optimal turn paths.At block 405, the present method 400 includes instructing the vehicle 101 to follow the planned trajectory 105 to pass through the intersection 135. The present method 400 may further include operating the vehicle 101 to follow the planned trajectory 105 by controlling or adjusting the steering, pushing, and braking inputs of the vehicle 101, or a combination thereof. The present method 400 may further include monitoring a lane 154 of the vehicle 101 as it passes the intersection 135, determining whether the lane 154 deviates from the planned trajectory 105, and in response to determining that the lane 154 deviates from the planned trajectory 105, generating an updated trajectory for the vehicle 101 to pass the intersection 135 based on the historical turn paths 303 and 305 of the human-driven vehicles 301.It is noted that the terms "substantially" and "about" may be used herein to represent the inherent degree of uncertainty that may be attributed to any quantitative comparison, value, measurement, or other representation. These terms are also used herein to represent the degree to which a quantitative representation may deviate from a specified reference without resulting in a change in the basic function of the subject matter in question.Although certain embodiments have been illustrated and described herein, various other changes and modifications may be made without departing from the spirit and scope of the claimed subject matter. Although various aspects of the claimed subject matter have been described herein, these aspects need not be used in combination. It is therefore intended that the appended claims cover all such changes and modifications as fall within the scope of the claimed subject matter.

Claims

An apparatus for trajectory planning, comprising at least one processor operable to: receive an instruction to turn within an intersection; plan a trajectory for an autonomous vehicle to turn within the intersection based on map data comprising at least one optimal turn path associated with the intersection, wherein the at least one optimal turn path is generated based on turn paths of at least one human-driven vehicle; and instruct the autonomous vehicle to follow the trajectory to pass through the intersection.The apparatus of claim 1, wherein the at least one optimal turn path is further generated based on historical sensor data of the human-driven vehicles over time.The apparatus of claim 2, wherein the trajectory is generated using a trained first machine learning algorithm based on the at least one optimal turn path and autonomous vehicle parameters versus human driven vehicle parameters associated with the at least one optimal turn path.The apparatus of claim 3, wherein the autonomous vehicle parameters and the human-driven vehicle parameters comprise vehicle length, minimum turning radii, steering system, acceleration and deceleration performance, or a combination thereof.The apparatus of claim 2, wherein the sensor data comprises speed data, acceleration data, steering data, yaw data, wheel slip data, lane departure data, time, weather conditions, vehicle type, vehicle size, minimum and maximum vehicle turn radii, or a combination thereof.The apparatus of claim 1, wherein the at least one optimal turn path is generated by a second trained machine learning algorithm configured to reduce path lengths of the optimal turn paths.The apparatus of claim 1, wherein the trajectory includes an imminent turn path and speed, time, and kinematics of the autonomous vehicle associated with the imminent turn path.The apparatus of claim 7, wherein the imminent turn path is one of the at least one optimal turn path.The apparatus of claim 1, wherein the at least one processor is further operable to operate the autonomous vehicle to follow the trajectory by controlling or adjusting steering, thrust, brake inputs of the autonomous vehicle, or a combination thereof.The apparatus of claim 1, wherein the at least one processor is further operable to: monitor a lane of the autonomous vehicle during a pass of the intersection; determine whether the lane deviates from the trajectory; and in response to determining that the lane deviates from the trajectory, generate an updated trajectory for the autonomous vehicle to pass the intersection based on historical turn paths of the human-driven vehicles.The apparatus of claim 1, wherein the map data includes pedestrian crossings, traffic lights, traffic signs, obstacles, road lanes, road edges, bankette, dividers, color markers, posts, or a combination thereof.The apparatus of claim 1, wherein the map data is high resolution map data or standard resolution map data.A method of trajectory planning, comprising: receiving an instruction to turn within an intersection; planning a trajectory for an autonomous vehicle to turn within the intersection based on map data comprising at least one optimal turn path associated with the intersection, wherein the at least one optimal turn path is generated based on turn paths of at least one human-driven vehicle; and instructing the autonomous vehicle to follow the trajectory to pass through the intersection.The method of claim 13, wherein: the at least one optimal turn path is further generated based on historical sensor data of the human-driven vehicles over time; and the sensor data comprises speed data, acceleration data, steering data, yaw data, wheel slip data, lane departure data, time of day, weather conditions, vehicle type, vehicle size, minimum and maximum vehicle turn radii, or a combination thereof.The method of claim 13, wherein: the trajectory is generated using a trained first machine learning algorithm based on the at least one optimal turn path and autonomous vehicle parameters versus human-driven vehicle parameters associated with the at least one optimal turn path; and the autonomous vehicle parameters and the human-driven vehicle parameters comprise vehicle length, minimum turn radii, steering system, acceleration and deceleration performance, or a combination thereof.The method of claim 13, wherein the at least one optimal turn path is generated by a second trained machine learning algorithm configured to reduce path lengths of the optimal turn paths.The method of claim 13, wherein the trajectory includes an imminent turn path and speed, time, and kinematics of the autonomous vehicle associated with the imminent turn path.The method of claim 13, wherein the method further comprises operating the autonomous vehicle to follow the trajectory by controlling or adjusting steering, thrust, brake inputs of the autonomous vehicle, or a combination thereof.The method of claim 13, wherein the method further comprises: monitoring a lane of the autonomous vehicle during a pass of the intersection; determining whether the lane deviates from the trajectory; and in response to determining that the lane deviates from the trajectory, generating an updated trajectory for the autonomous vehicle to pass the intersection based on historical turn paths of the human-driven vehicles.The method of claim 13, wherein the map data is high resolution map data or standard resolution map data.

Images