Learning-based controller for autonomous driving
By using a learning-based model to predict the controller and adjust weights online to generate optimized control commands, this technology addresses the problem of failing to consider changes in the vehicle's environment and state in existing technologies, thereby improving the safety and driving experience of autonomous vehicles.
Patent Information
- Application Number
- CN202011030432.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2020-09-27
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2040-09-27
AI Technical Summary
Existing model predictive controllers fail to effectively account for changes in the vehicle environment and physical state in autonomous vehicles, leading to a decline in safety and driving experience.
A learning-based model predictive controller is adopted to generate optimized control commands by adjusting the weights of the cost function online, taking into account the vehicle's dynamic environment and physical state, and periodically updating the control strategy.
It improves the safety and driving experience of autonomous vehicles by dynamically adjusting control strategies to adapt to changes in the environment and physical state, thereby optimizing vehicle behavior.
Smart Images

Figure CN113495559B_ABST
Abstract
Description
Technical Field
[0001] Embodiments of the present disclosure generally relate to operating an autonomous vehicle. More specifically, embodiments of the present disclosure relate to a learning-based controller for autonomous driving. Background Art
[0002] A vehicle operating in an autonomous mode (e.g., without a driver) can relieve occupants, particularly the driver, of certain driving-related responsibilities. When operating in an autonomous mode, the vehicle can navigate to various locations using onboard sensors, allowing the vehicle to travel with minimal human interaction or, in some cases, without any passengers.
[0003] A vehicle controller of an autonomous vehicle may generate control commands to move the vehicle according to a desired path or route. The controller may utilize a static control algorithm.
[0004] The model predictive controller can generate a sequence of commands to be applied at a future time frame to move the controlled object along the predicted path. The command sequence can be optimized for different terms, such as lateral trajectory error, driving direction error, and sudden changes in speed, acceleration, driving direction, etc. The first one in the command sequence is applied to the controlled object. At a subsequent time (e.g., the next cycle), the process is repeated, and the first one in the new command sequence is applied to the controlled object at each cycle.
[0005] Such controllers can be used to control and operate an autonomous vehicle (ADV) to travel along a target path at a target speed. MPC (Model Predictive Control) can use static optimization algorithms and a static vehicle model to generate optimized command sequences. However, these control algorithms may not take into account changes in the vehicle environment or changes in the vehicle. No matter what the vehicle's environment is and how much the vehicle's physical state has changed, the controller will use the same algorithm to achieve the desired speed and steering behavior. This can reduce safety and cause an uncomfortable driving experience. Summary of the invention
[0006] Embodiments of the present disclosure provide a method for operating an autonomous driving vehicle ADV, a non-transitory machine-readable medium storing instructions, and a data processing system.
[0007] In one aspect of the present disclosure, a method for operating an autonomous driving vehicle ADV includes: generating a control command using a model predictive control (MPC) controller, the MPC controller including an optimizer using a cost function, the cost function being configured with one or more weights associated with a cost term of the cost function; applying the control command to a dynamic model of the ADV to simulate the behavior of the ADV; adjusting the one or more weights based on an evaluation of the dynamic model in response to the control command to obtain an adjusted cost function of the MPC controller; and generating a second control command for achieving movement of the ADV using the MPC controller having the adjusted cost function.
[0008] In another aspect of the present disclosure, a non-transitory machine-readable medium storing instructions is disclosed, which, when executed by a processor, causes the processor to perform operations for operating an autonomous driving vehicle ADV, the operations including: generating a control command using a model predictive control (MPC) controller, the MPC controller including an optimizer using a cost function, the cost function being configured with one or more weights associated with a cost term of the cost function; applying the control command to a dynamic model of the ADV to simulate the behavior of the ADV; adjusting the one or more weights based on an evaluation of the dynamic model in response to the control command to obtain an adjusted cost function of the MPC controller; and generating a second control command for achieving movement of the ADV using the MPC controller having the adjusted cost function.
[0009] In another aspect of the present disclosure, a data processing system includes: a processor; and a memory connected to the processor to store instructions, which, when executed by the processor, cause the processor to perform operations for operating an autonomous driving vehicle ADV, the operations including: generating a control command using a model predictive control (MPC) controller, the MPC controller including an optimizer using a cost function, the cost function being configured with one or more weights associated with a cost term of the cost function; applying the control command to a dynamic model of the ADV to simulate the behavior of the ADV; adjusting the one or more weights based on an evaluation of the dynamic model in response to the control command to obtain an adjusted cost function of the MPC controller; and generating a second control command for achieving movement of the ADV using the MPC controller having the adjusted cost function. BRIEF DESCRIPTION OF THE DRAWINGS
[0010] Embodiments of the present disclosure are illustrated by way of example and not by way of limitation in the figures of the accompanying drawings in which like references indicate similar elements.
[0011] Figure 1is a block diagram illustrating a networked system according to one embodiment.
[0012] Figure 2 is a block diagram illustrating an example of an automated vehicle according to one embodiment.
[0013] FIG. 3A to FIG. 3B is a block diagram illustrating an example of a perception and planning system for use with an automated vehicle, according to one embodiment.
[0014] Figure 4 A block diagram showing a system architecture for autonomous driving according to one embodiment.
[0015] Figure 5 A process for controlling an autonomous vehicle based on a scenario according to one embodiment is shown.
[0016] Figure 6 A learning-based model predictive controller according to one embodiment is shown.
[0017] Figure 7 A system for controlling an autonomous vehicle based on a learning-based model predictive controller is shown according to one embodiment. DETAILED DESCRIPTION
[0018] Various embodiments and aspects of the present disclosure will be described with reference to the details discussed below, and the accompanying drawings will illustrate various embodiments. The following description and the accompanying drawings are examples of the present disclosure and should not be construed as limiting the present disclosure. Many specific details are described herein to provide a thorough understanding of various embodiments of the present disclosure. However, in some cases, in order to provide a concise discussion of embodiments of the present disclosure, known or conventional details are not described.
[0019] Reference to "one embodiment" or "an embodiment" in the specification means that a particular feature, structure or characteristic described in conjunction with the embodiment may be included in at least one embodiment of the present disclosure. The expression "in one embodiment" appearing in various places in the specification does not necessarily refer to the same embodiment.
[0020] According to some embodiments, an online learning system updates parameters for a controller (e.g., a model predictive controller) based on the current physical environment of an autonomous vehicle (ADV) and the current physical state of the ADV. The MPC can generate a series of optimized control commands based on the predicted movements of the ADV. These predicted movements will follow a target path or route (including the location of waypoints along the route and the direction of travel), while various items are optimized to reduce undesirable conditions (e.g., large changes in speed, acceleration, and direction of travel). Therefore, the control commands are optimized by the MPC and, when applied to the ADV control actuators, will cause the ADV to follow the target path.
[0021] These control commands are then applied to a vehicle dynamics model representing the ADV to simulate how the ADV responds. In the simulation, real-time environmental conditions are considered, such as traffic flow, road conditions (e.g., wet, slippery, icy), sensed obstacles, etc. The ADV may also have passengers or other cargo items that increase the mass of the ADV. The fuel, battery state of charge, or battery health may also change. These changes may be considered in the simulation. The optimization parameters (e.g., weights or coefficients) of the MPC controller are evaluated and adjusted based on the simulation.
[0022] The adjusted MPC controller is used to generate one or more control commands that take into account the current environmental conditions of the ADV. This adjustment can be performed periodically, for example, based on each driving cycle. In this way, the MPC controller is periodically adjusted "online" while driving. Other aspects and details are discussed in the present disclosure.
[0023] Figure 1 is a block diagram showing an automatic vehicle network configuration according to one embodiment of the present disclosure. Figure 1 , the network configuration 100 includes an automatic vehicle 101 that can be communicatively connected to one or more servers 103 to 104 via a network 102. Although one automatic vehicle is shown, multiple automatic vehicles can be connected to each other and / or to the servers 103 to 104 via the network 102. The network 102 can be any type of network, wired or wireless, such as a local area network (LAN), a wide area network (WAN) such as the Internet, a cellular network, a satellite network, or a combination thereof. The one or more servers 103 to 104 can be any type of server or server cluster, such as a web or cloud server, an application server, a back-end server, or a combination thereof. The servers 103 to 104 can be data analysis servers, content servers, traffic information servers, maps and points of interest (MPOI) servers, or location servers, etc.
[0024] An autonomous vehicle refers to a vehicle that can be configured in an autonomous mode in which the vehicle navigates through an environment with little or no input from a driver. Such an autonomous vehicle may include a sensor system having one or more sensors configured to detect information about the environment in which the vehicle is operating. The vehicle and its associated one or more controllers use the detected information to navigate through the environment. The autonomous vehicle 101 may operate in a manual mode, a fully autonomous mode, or a partially autonomous mode.
[0025] In one embodiment, the automatic vehicle 101 includes, but is not limited to, a perception and planning system 110, a vehicle control system 111, a wireless communication system 112, a user interface system 113, and a sensor system 115. The automatic vehicle 101 may also include certain common components included in a normal vehicle, such as an engine, wheels, a steering wheel, a transmission, etc., which may be controlled by the perception and planning system 110 and / or the vehicle control system 111 using various communication signals and / or commands, such as acceleration signals or commands, deceleration signals or commands, steering signals or commands, braking signals or commands, etc.
[0026] Components 110 to 115 may be communicatively coupled to one another via an interconnect, a bus, a network, or a combination thereof. For example, components 110 to 115 may be communicatively coupled to one another via a controller area network (CAN) bus. The CAN bus is a vehicle bus standard designed to allow microcontrollers and devices to communicate with one another in applications without a host computer. The CAN bus is a message-based protocol originally designed for multi-way wiring within an automobile, but is also used in many other environments.
[0027] Reference now Figure 2 In one embodiment, the sensor system 115 includes, but is not limited to, one or more cameras 211, a global positioning system (GPS) unit 212, an inertial measurement unit (IMU) 213, a radar unit 214, and a light detection and ranging (LIDAR) unit 215. The GPS system 212 may include a transceiver that is operable to provide information about the location of the automatic vehicle. The IMU unit 213 may sense the position and direction changes of the automatic vehicle based on inertial acceleration. The radar unit 214 may represent a system that uses radio signals to sense objects within the local environment of the automatic vehicle. In some embodiments, in addition to sensing objects, the radar unit 214 may also sense the speed and / or direction of travel of the object. The LIDAR unit 215 may use lasers to sense objects in the environment in which the automatic vehicle is located. The LIDAR unit 215 may include one or more laser sources, a laser scanner, and one or more detectors, among other system components. The camera 211 may include one or more devices for capturing images of the environment surrounding the automatic vehicle. The camera 211 may be a still camera and / or a video camera. For example, by mounting the camera on a rotating and / or tilting platform, the camera may be capable of mechanical movement.
[0028] The sensor system 115 may also include other sensors, such as sonar sensors, infrared sensors, steering sensors, throttle sensors, brake sensors, and audio sensors (e.g., microphones). The audio sensor may be configured to capture sound from the environment surrounding the automated vehicle. The steering sensor may be configured to sense the steering angle of the steering wheel, the wheels of the vehicle, or a combination thereof. The throttle sensor and the brake sensor sense the throttle position and the brake position of the vehicle, respectively. In some cases, the throttle sensor and the brake sensor may be integrated into an integrated throttle / brake sensor.
[0029] In one embodiment, the vehicle control system 111 includes, but is not limited to, a steering unit 201, a throttle unit 202 (also referred to as an acceleration unit), and a brake unit 203. The steering unit 201 is used to adjust the direction or driving direction of the vehicle. The throttle unit 202 is used to control the speed of the motor or engine, which in turn controls the speed and acceleration of the vehicle. The brake unit 203 decelerates the vehicle by providing friction to slow down the speed of the wheels or tires of the vehicle. Note that, as Figure 2 The components shown may be implemented in hardware, software or a combination thereof.
[0030] Back to Figure 1 , the wireless communication system 112 will allow communication between the automatic vehicle 101 and external systems, such as devices, sensors, other vehicles, etc. For example, the wireless communication system 112 can communicate wirelessly with one or more devices, such as servers 103-104 on the network 102, directly or via a communication network. The wireless communication system 112 can communicate with another component or system using any cellular communication network or wireless local area network (WLAN), such as using WiFi. The wireless communication system 112 can communicate directly with devices (e.g., a passenger's mobile device, a display device, a speaker within the vehicle 101) using, for example, an infrared link, Bluetooth, etc. The user interface system 113 can be part of a peripheral device implemented within the vehicle 101, including, for example, a keyboard, a touch screen display device, a microphone, and a speaker.
[0031] Some or all of the functions of the automated vehicle 101 may be controlled or managed by the perception and planning system 110, particularly when operating in an automated driving mode. The perception and planning system 110 includes the necessary hardware (e.g., one or more processors, memory, storage devices) and software (e.g., operating system, planning and routing programs) to receive information from the sensor system 115, the control system 111, the wireless communication system 112, and / or the user interface system 113, process the received information, plan a route or path from a starting point to a destination point, and then drive the vehicle 101 based on the planning and control information. Alternatively, the perception and planning system 110 may be integrated with the vehicle control system 111.
[0032] For example, a user who is a passenger may specify a starting location and a destination for a trip, for example, via a user interface. The perception and planning system 110 obtains trip-related data. For example, the perception and planning system 110 may obtain location and route information from an MPOI server, which may be part of the servers 103 to 104. The location server provides location services, and the MPOI server provides map services and POIs for specific locations. Alternatively, such location and MPOI information may be cached locally in a permanent storage device of the perception and planning system 110.
[0033] As the automated vehicle 101 moves along the route, the perception and planning system 110 may also obtain real-time traffic information from a traffic information system or server (TIS). Note that the servers 103 to 104 may be operated by a third-party entity. Alternatively, the functionality of the servers 103 to 104 may be integrated with the perception and planning system 110. Based on the real-time traffic information, MPOI information and location information, and real-time local environmental data (e.g., obstacles, objects, nearby vehicles) detected or sensed by the sensor system 115, the perception and planning system 110 may, for example, plan an optimal route via the control system 111 and drive the vehicle 101 to reach a designated destination safely and efficiently.
[0034] The server 103 may be a data analysis system that performs data analysis services for various clients. In one embodiment, the data analysis system 103 includes a data collector 121 and a machine learning engine 122. The data collector 121 collects driving statistics 123 from various vehicles, which are automatic vehicles or conventional vehicles driven by human drivers. The driving statistics 123 include information indicating the driving commands (e.g., throttle, brake, steering commands) and responses (e.g., speed, acceleration, deceleration, direction) of the vehicle captured by the vehicle's sensors at different points in time. The driving statistics 123 may also include information describing the driving environment at different points in time, such as routes (including starting and destination locations), MPOIs, road conditions, weather conditions, etc.
[0035] Based on the driving statistics 123, the machine learning engine 122 generates or trains a set of rules, algorithms, and / or predictive models 124 for various purposes. In one embodiment, the algorithms 124 may include algorithms used by the model predictive controller of the present disclosure. The algorithms 124 may then be uploaded to the ADV to be utilized in real time during autonomous driving.
[0036] Figure 3A and Figure 3B is a block diagram illustrating an example of a perception and planning system for use with an autonomous vehicle according to one embodiment. System 300 may be implemented as Figure 1A portion of an automated vehicle 101 includes, but is not limited to, a perception and planning system 110, a control system 111, and a sensor system 115. FIG. 3A to FIG. 3B The perception and planning system 110 includes but is not limited to a positioning module 301 , a perception module 302 , a prediction module 303 , a decision module 304 , a planning module 305 , a control module 306 , a routing module 307 and a learning-based MPC 308 .
[0037] Some or all of the modules 301 to 308 may be implemented in software, hardware, or a combination thereof. For example, these modules may be installed in permanent storage device 352, loaded into memory 351, and executed by one or more processors (not shown). Note that some or all of these modules may be communicatively coupled to Figure 2 Some or all modules of the vehicle control system 111, or Figure 2 Some or all modules of the vehicle control system 111 are integrated. Some of the modules 301 to 308 can be integrated together as an integrated module.
[0038] The positioning module 301 determines the current location of the automatic vehicle 300 (e.g., using the GPS unit 212) and manages any data related to the user's trip or route. The positioning module 301 (also known as the map and route module) manages any data related to the user's trip or route. The user can log in, for example, via a user interface and specify the starting location and destination of the trip. The positioning module 301 communicates with other components of the automatic vehicle 300, such as map and route information 311, to obtain trip-related data. For example, the positioning module 301 can obtain location and route information from a location server and a map and POI (MPOI) server. The location server provides location services, and the MPOI server provides map services and POIs at specific locations, which can be cached as part of the map and route information 311. When the automatic vehicle 300 moves along the route, the positioning module 301 can also obtain real-time traffic information from a traffic information system or server.
[0039] Based on the sensor data provided by the sensor system 115 and the positioning information obtained by the positioning module 301, the perception of the surrounding environment is determined by the perception module 302. The perception information may represent what an average driver would perceive around the vehicle that the driver is driving. The perception may include lane configuration, traffic light signals, relative positions of other vehicles, pedestrians, buildings, sidewalks, or other traffic-related signs (e.g., stop signs, give way signs), etc., for example, in the form of objects. The lane configuration includes information describing one or more lanes, such as the shape of the lane (e.g., straight or curvature), the width of the lane, how many lanes there are in the road, one-way or two-way lanes, merging or splitting lanes, exit lanes, etc.
[0040] The perception module 302 may include a computer vision system or functionality of a computer vision system to process and analyze images captured by one or more cameras to identify objects and / or features in the environment of the automated vehicle. Objects may include traffic signs, road boundaries, other vehicles, pedestrians, and / or obstacles, among others. The computer vision system may use object recognition algorithms, video tracking, and other computer vision techniques. In some embodiments, the computer vision system may map the environment, track objects, and estimate the speed of objects, among others. The perception module 302 may also detect objects based on other sensor data provided by other sensors such as radar and / or LIDAR.
[0041] For each object, the prediction module 303 predicts the behavior that the object will exhibit in the environment. The prediction is performed based on the perception data that perceives the driving environment at a point in time according to a set of map / route information 311 and traffic rules 312. For example, if the object is a vehicle in the opposite direction and the current driving environment includes an intersection, the prediction module 303 will predict whether the vehicle is likely to move directly forward or turn. If the perception data indicates that there is no traffic light at the intersection, the prediction module 303 may predict that the vehicle may have to come to a complete stop before entering the intersection. If the perception data indicates that the vehicle is currently in a left turn lane or a right turn lane, the prediction module 303 may predict that the vehicle will be more likely to make a left turn or a right turn, respectively.
[0042] For each object, the decision module 304 makes a decision about how to handle the object. For example, for a particular object (e.g., another vehicle in an intersection) and its metadata describing the object (e.g., speed, direction, steering angle), the decision module 304 decides how to encounter the object (e.g., overtake, yield, stop, pass). The decision module 304 may make such decisions based on a set of rules, such as traffic rules or driving rules 312, which may be stored in the persistent storage device 352.
[0043] The routing module 307 is configured to provide one or more routes or paths from the starting point to the destination point. For a given trip from the starting position to the destination position, for example, received from a user, the routing module 307 obtains route and map information 311 and determines all possible routes or paths from the starting position to the destination position. The routing module 307 can generate a reference line in the form of a topographic map for each route it determines to reach the destination position from the starting position. The reference line refers to an ideal route or path without any interference from other vehicles, obstacles or traffic conditions. That is, if there are no other vehicles, pedestrians or obstacles on the road, the ADV should follow the reference line accurately or closely. The topographic map is then provided to the decision module 304 and / or the planning module 305. The decision module 304 and / or the planning module 305 consider other data provided by other modules, such as traffic conditions from the positioning module 301, the driving environment perceived by the perception module 302, and traffic conditions predicted by the prediction module 303, and check all possible routes to select and modify one of the best routes. The actual path or route used to control the ADV may be close to or different from the reference line provided by the routing module 307, depending on the particular driving environment at that point in time.
[0044] Based on the decision of each of the perceived objects, the planning module 305 uses the reference line provided by the routing module 307 as a basis to plan the path or route of the automatic vehicle and the driving parameters (e.g., distance, speed and / or steering angle). That is, for a given object, the decision module 304 decides what to do with the object, and the planning module 305 decides how to do it. For example, for a given object, the decision module 304 may decide to pass through the object, and the planning module 305 may determine whether to pass on the left or right side of the object. Planning and control data are generated by the planning module 305, including information describing how the vehicle 300 will move in the next movement cycle (e.g., the next route / path segment). For example, the planning and control data may indicate that the vehicle 300 moves at a speed of 30 miles per hour (mph) for 10 meters and then changes to the right lane at a speed of 25mph.
[0045] Based on the planning and control data, the control module 306 controls and drives the automated vehicle according to the route or path defined by the planning and control data by sending appropriate commands or signals to the vehicle control system 111. The planning and control data includes sufficient information to drive the vehicle from a first point on the route or path to a second point using appropriate vehicle settings or driving parameters (e.g., throttle, brake, steering commands) at different points in time along the route or path.
[0046] In one embodiment, the planning phase is performed in multiple planning cycles, also referred to as driving cycles, for example, in each time interval of 100 milliseconds (ms). For each planning cycle or driving cycle, one or more control commands will be issued based on the planning and control data. That is, for every 100ms, the planning module 305 plans the next route segment or path segment, for example, including the target position and the time required for the ADV to reach the target position. Alternatively, the planning module 305 may further specify a specific speed, direction and / or steering angle, etc. In one embodiment, the planning module 305 may plan a route segment or path segment for the next predetermined time period such as 5 seconds. For each planning cycle, the planning module 305 plans the target position of the current cycle (e.g., the next 5 seconds) based on the target position planned in the previous cycle. Then, the control module 306 generates one or more control commands (e.g., throttle, brake, steering control commands) based on the planning and control data of the current cycle.
[0047] Note that the decision module 304 and the planning module 305 may be integrated as an integrated module. The decision module 304 / planning module 305 may include a navigation system or functionality of a navigation system to determine a travel path for the automated vehicle. For example, the navigation system may determine a series of speeds and directional travel directions to affect movement of the automated vehicle along a path that substantially avoids perceived obstacles while generally advancing the automated vehicle along a road-based path to a final destination. The destination may be set based on user input via the user interface system 113. The navigation system may dynamically update the travel path while the automated vehicle is operating. The navigation system may combine data from a GPS system and one or more maps to determine a travel path for the automated vehicle.
[0048] The learning-based MPC 308 includes an MPC controller having an optimizer and a vehicle model. The optimizer can use the cost function and the vehicle model to generate a sequence of control commands (e.g., throttle, steering, and / or braking commands) that follow the vehicle path along the target vehicle trajectory. These commands are generated while optimizing for different cost terms (e.g., lateral trajectory error, driving direction error, speed, steering, acceleration, steering rate of change, and / or acceleration rate of change). Each cost term can be represented in the cost function to reduce undesirable behavior. A weight can be associated with each term and applied to each term (e.g., multiplication) to modify the impact of each term on the total computational cost.
[0049] The learning-based MPC module can be an online learning module, which means that it continuously updates the MPC controller based on new data (e.g., vehicle environment and physical parameters) while the vehicle is driving, for example. The learning-based MPC 308 can be integrated with another module, such as the planning module 305 and / or the control module 306.
[0050] A general example of a cost function for an MPC controller is shown below, where J is the total computational cost, w x is the weight corresponding to term x (x=1, 2, ...), and N is the point along the target trajectory of ADV.
[0051]
[0052] These terms can be optimized by minimizing the computational cost J. These terms can include at least one of the following: lateral trajectory error (constrained for the distance of the ADV from the target trajectory), driving direction error (constrained for the error between the ADV driving direction and the target trajectory direction at a certain point), speed cost (constrained for speed changes), steering cost (constrained for steering changes), acceleration cost (constrained for acceleration changes), steering rate of change (constrained for how fast the steering changes), braking cost (constrained for braking) and acceleration rate of change (constrained for how fast the acceleration changes). In some embodiments, the cost function includes at least two of the above terms. In other embodiments, the cost function includes all of the above terms. While considering the above terms, a sequential control command (throttle, steering, braking) can be generated to optimally follow the target trajectory.
[0053] These terms are generated by the MPC optimizer using the static model of the ADV. The MPC predicts how the static model will move along the trajectory in response to different control commands to determine the optimized control command sequence based on the static model of the ADV and minimize the cost function.
[0054] MPC itself does not take into account environmental factors or current physical states in the ADV, both of which may change as the vehicle travels along the trajectory to the intended destination. The vehicle model of the MPC controller may be a simplified vehicle model that includes the vehicle's dimensions and mass. The vehicle model may be static, meaning that it is constant over time (e.g., from one frame to another).
[0055] The learning-based MPC controller 308 feeds these control commands generated using the static MPC controller into a vehicle dynamics model that includes the vehicle mass and geometry (e.g., shape and size), but may also include additional dynamic parameters that model how the ADV responds to different control commands and how the ADV responds under different environmental conditions of the ADV and under different physical states of the ADV (e.g., full tank, empty tank, low battery, wet brakes, etc.). The dynamic model also more accurately determines how the ADV responds to specific commands (e.g., at higher speeds vs. lower speeds) and larger steering angles. The dynamic vehicle model can be generated by actually testing a physical ADV to more accurately represent how the ADV will behave under different speeds, road conditions, steering angles, with different passengers and cargo loads, etc.
[0056] The learning-based MPC 308 adjusts the model predictive controller based on evaluating the behavior of the vehicle dynamics model in response to the control commands. The weights of the MPC can be adjusted. The adjusted model predictive controller can then be used to generate control commands for the physical ADV. This can be repeated periodically, for example, based on each driving cycle. More aspects and details will be discussed in other sections.
[0057] Figure 4 400 is a block diagram showing a system architecture for autonomous driving according to one embodiment. Figure 3A and Figure 3B The system architecture of the autonomous driving system is shown in Figure 2. Figure 4 , the system architecture 400 includes, but is not limited to, an application layer 401, a planning and control (PNC) layer 402, a perception layer 403, a driver layer 404, a firmware layer 405, and a hardware layer 406. The application layer 401 may include a user interface or configuration application that interacts with a user or passenger of the autonomous vehicle, for example, functionality associated with the user interface system 113. The PNC layer 402 may include functionality of at least the planning module 305 and the control module 306. The perception layer 403 may include functionality of at least the perception module 302. In one embodiment, there are additional layers that include functionality of the prediction module 303 and / or the decision module 304. Alternatively, such functionality may be included in the PNC layer 402 and / or the perception layer 403. The system architecture 400 also includes a driver layer 404, a firmware layer 405, and a hardware layer 406. The firmware layer 405 may represent at least functionality of the sensor system 115, which may be implemented in the form of a field programmable gate array (FPGA). The hardware layer 406 may represent hardware of the autonomous vehicle, such as the control system 111. The layers 401 to 403 may communicate with the firmware layer 405 and the hardware layer 406 via the device driver layer 404 .
[0058] Figure 5 A process 500 of controlling an ADV using a learning-based model predictive controller according to some embodiments is shown. Operation 501 includes generating control commands using an MPC controller including a cost function having one or more weights associated with a cost term of the cost function. The control commands are also not adjustable based on current environmental conditions and physical properties of the ADV.
[0059] Operation 502 includes applying control commands to a dynamic model of the ADV to simulate the behavior of the ADV. Although the MPC controller used in operation 501 may include a simplified model of the ADV, the dynamic model may more accurately model the behavior of the ADV by taking into account more sophisticated and nonlinear behavior of the ADV, which may be determined through testing and experimentation.
[0060] In addition, although not required, the simulation of the dynamic model can be performed in a virtual environment that can include objects and structures currently sensed around the ADV to take into account the current environment of the ADV when generating control commands. The virtual environment can include a two-dimensional or three-dimensional representation of the current environment around the ADV. Although simplified, the environment can include the boundaries of objects (e.g., pedestrians, vehicles, structures) and the geometry of the road boundaries. The virtual environment can be generated based on the sensed data from the sensor system 115, and / or information from the map and route information 311, the positioning module 301, and other modules of the perception and planning system 110.
[0061] Operation 503 includes adjusting one or more weights based on an evaluation of the dynamic model in response to the control command, thereby generating an adjusted cost function for the MPC controller. For example, the operation may determine one or more scores associated with the one or more weights based on an evaluation of the dynamic model in response to the control command. The evaluation may also be based solely on one or more environmental conditions around the ADV, such as moisture, traffic flow, etc. As discussed, the MPC controller used at box 501 may include a cost function having items that are optimized when tracking the ADV along a target trajectory. Each of these items may have an associated weight for adjusting the importance of each item (e.g., relative to another). Each item may be evaluated (e.g., scored) based on how the dynamic model performs in a simulated environment.
[0062] For example, if the ambient conditions are wet (which makes the ADV more slippery than usual), the steering terms may be evaluated as unfavorable (low score) to discourage steering changes. Accordingly, the weights associated with these terms may then be adjusted to increase to further constrain steering in wet conditions. Here, the adjusted MPC will generate new control commands for less active steering.
[0063] In another example, analysis of the dynamic model in a simulated environment may show that the dynamic model is moving too close to a sensed object, such as a curb, pedestrian, or other vehicle. Again, the terms are evaluated, and the weights for lateral trajectory error and heading error may be increased to further constrain these errors so that the ADV follows the target trajectory more aggressively. When the object is no longer sensed, these weights may be reduced (e.g., in a subsequent cycle) to relax ADV control and improve ADV ride comfort.
[0064] In another example, the ADV may sense that traffic flow is low. Under these conditions, the weight associated with the speed term may be reduced so that speed increases are less constrained. Thus, when traffic flow is low, the learning-based MPC may increase speed more aggressively (e.g., toward a speed limit constraint). Conversely, when traffic flow is high, the weight associated with the speed cost term may be increased to further constrain faster speeds.
[0065] In another example, when traffic volume is high, the weight associated with braking may be reduced to reduce the restriction on braking. It should be understood that although "high" traffic volume and "low" traffic volume are relative terms; however, thresholds or other known mechanisms may be implemented to infer whether traffic volume is low or high.
[0066] In another example, the ADV undergoes physical changes while driving. For example, the brakes can get hot or wet, changing the vehicle's response to braking commands. The ADV can carry passengers or cargo that change the mass of the ADV. The amount of gasoline in the ADV tank, or the charge state or health state of the electric ADV can also change the way the vehicle responds to throttle, steering, and braking commands. When the ADV undergoes physical changes, the same results can lead to different ADV behaviors. Changes in the physical properties of the ADV can be sensed by the ADV's sensor system and used to update the ADV's dynamic model and / or for directly evaluating the weights of cost items. Therefore, the dynamic model and / or weights can represent the current state of the ADV.
[0067] Operation 504 includes generating, using the MPC controller with the adjusted cost function, a second control command for achieving movement of the ADV. In this manner, the learning-based MPC controller uses a control algorithm optimized for the current physical properties of the ADV and the current environment of the ADV. It should be understood that the current environment of the ADV includes an effective environment of the ADV, for example, within an effective perimeter around the ADV associated with the control command of the ADV. This may be, for example, 10 meters, 20 meters, 30 meters, 40 meters, etc. Such a perimeter may vary depending on the application and may be determined through routine testing and experimentation.
[0068] Figure 6A learning based MPC module 308 is shown according to some embodiments. References are provided to the model predictive control module 602. The references may include, for example, a target ADV trajectory, a current driving direction, and a current speed of the ADV.
[0069] The MPC module 602 may include a vehicle model 670 and a cost function 674. The cost function may include a cost term 678 and an associated weight 676. The MPC module may generate a future command sequence 672 (e.g., throttle, brake, and steering) that will predictively affect the movement of the vehicle model so that the vehicle model travels according to the reference while minimizing the cost function.
[0070] The first in the sequence of commands is taken and applied to the vehicle dynamics model 606 to simulate a more accurate representation of the vehicle's behavior than that generated at the MPC module 602. As discussed, the vehicle dynamics model 606 may include the dynamic response of the ADV, which may be determined based on testing of an actual ADV. Such a response may be non-linear and more accurately reflect the dynamic behavior of the ADV than the vehicle model 670 of the MPC module 602. As described above, the vehicle dynamics model may also include physical properties of the ADV that may change during driving (e.g., brake wetness, fuel state, battery state, passengers, and cargo). The vehicle dynamics model may adjust the response to the control commands based on these physical properties.
[0071] The environment module 608 may collect information describing the environmental conditions 609 surrounding the current ADV, such as pedestrians, other vehicles, road boundaries or buildings, road conditions (wet, slippery, icy, snowy), weather conditions, and traffic volume.
[0072] The parameter evaluation module 610 can evaluate the vehicle dynamics behavior relative to a reference, for example, to see how well the vehicle dynamics model tracks the trajectory. The vehicle dynamics model can generate different trajectories based on the predicted trajectory generated by the MPC module 602 because the vehicle model is different. At the parameter evaluation module 610, the response behavior of the dynamics model to the control commands can be evaluated for the current environmental conditions.
[0073] In some embodiments, the evaluation of the dynamic model in response to the control command is based on at least one of: the proximity of the dynamic model to a pedestrian, another vehicle, or a building; the speed of the dynamic model with reference to a speed constraint; the acceleration of the dynamic model relative to an acceleration constraint; the position of the dynamic model relative to a road boundary or path; a change in driving direction or speed that may cause ride discomfort; and control force. A score 611 may be generated for the weights of the cost function based on how well each term performs in the dynamic model. For example, if the predicted path of the dynamic model becomes too close to a pedestrian, the weights may be adjusted to increase the constraints on speed, acceleration, and acceleration increases.
[0074] At the configuration adjustment module, multiple permutations of one or more weights may be evaluated until the score reaches the desired range. Those one or more weights that produce scores within the desired range may serve as one or more weights of the adjusted cost function. At the new configuration generator 614, an adjusted cost function is generated using the adjusted weights. The adjusted cost function is then used as the effective cost function. In other words, the MPC controller with the adjusted cost function may then generate another control command.
[0075] The control command fed to the vehicle control may be the first of another sequence of control commands generated by the updated MPC controller. As described above, the MPC controller generates a sequence of control commands by optimizing a term (in this case, an adjusted cost function) by minimizing a cost function while satisfying a target control object. Here, the target control object travels along a reference trajectory.
[0076] Figure 7 3 shows an ADV system with a control module 306 that uses a learning-based MPC module to control the ADV. The planning module 305 can generate a path or route that the control module uses as a reference. The path or route can be determined by the planning module based on a reference line, as described in other sections. The path or route can be considered a target trajectory that the learning-based MPC module 602 will attempt to track.
[0077] As discussed, the learning-based MPC module can dynamically adjust (e.g., while the ADV is traveling) to take into account the real-time environmental conditions of the ADV. These environmental conditions can be collected from the servers 104 and 103 as well as the positioning module 301, the map and route information 311, the sensor system 115, and other modules. After adjusting to the environment and taking into account the physical properties of the ADV, the MPC module can generate and optimize control commands (e.g., throttle, steering, and / or braking) to be transmitted to the control system 111.
[0078] It should be understood that the MPC can generate throttle, steering and / or braking commands for a single drive cycle. In other words, multiple commands (for different control units) can be generated for a single drive cycle and transmitted to the corresponding control units (e.g., steering actuator, throttle actuator, brake actuator) to achieve corresponding and proportional movement of the ADV.
[0079] The sensor system 115 may feed back ADV information to the control module 306, such as current driving direction, current steering, current speed, etc. The learning-based MPC module may periodically self-adjust so that the control module generates control commands that take into account environmental factors surrounding the ADV and the physical properties of the ADV that may change over time.
[0080] Note that some or all of the components shown and described above may be implemented in software, hardware, or a combination thereof. For example, such components may be implemented as software installed and stored in a permanent storage device, which may be loaded into a memory by a processor (not shown) and executed in the memory to implement the processes or operations described throughout this application. Alternatively, such components may be implemented as executable code programmed or embedded into dedicated hardware (such as an integrated circuit (e.g., an application-specific integrated circuit or ASIC), a digital signal processor (DSP), or a field programmable gate array (FPGA)), which may be accessed via a corresponding driving program and / or operating system from an application. In addition, such components may be implemented as specific hardware logic in a processor or processor core as part of an instruction set that can be accessed by a software component through one or more specific instructions.
[0081] Some portions of the foregoing detailed description have been presented in terms of algorithms and symbolic representations of operations on data bits within a computer memory. These algorithmic descriptions and representations are the means used by those skilled in the data processing arts to most effectively convey the substance of their work to others skilled in the art. An algorithm is generally considered herein to be a self-consistent sequence of operations leading to a desired result. These operations are those requiring physical manipulation of physical quantities.
[0082] It should be borne in mind, however, that all of these and similar terms are intended to be associated with the appropriate physical quantities and are merely convenient labels applied to these quantities. Unless otherwise expressly indicated in the above discussion, it should be understood that throughout the specification, discussions utilizing terms such as those set forth in the appended claims refer to the actions and processes of a computer system or similar electronic computing device that manipulates data represented as physical (electronic) quantities within the computer system's registers and memories and transforms the data into other data similarly represented as physical quantities within the computer system's memories or registers or other such information storage, transmission, or display devices.
[0083] Embodiments of the present disclosure also relate to devices for performing the operations herein. Such a computer program is stored in a non-transitory computer-readable medium. A machine-readable medium includes any mechanism for storing information in a form readable by a machine (e.g., a computer). For example, a machine-readable (e.g., computer-readable) medium includes a machine (e.g., computer) readable storage medium (e.g., a read-only memory ("ROM"), a random access memory ("RAM"), a magnetic disk storage medium, an optical storage medium, a flash memory device).
[0084] The processes or methods depicted in the foregoing figures may be performed by processing logic including hardware (e.g., circuits, dedicated logic, etc.), software (e.g., embodied on a non-transitory computer-readable medium), or a combination of both. Although the processes or methods are described above in terms of some sequential operations, it should be understood that some of the operations may be performed in a different order. In addition, some operations may be performed in parallel rather than sequentially.
[0085] The embodiments of the present disclosure are not described with reference to any particular programming language. It will be appreciated that a variety of programming languages may be used to implement the teachings of the embodiments of the present disclosure as described herein.
[0086] In the above description, embodiments of the present invention have been described with reference to specific exemplary embodiments of the present disclosure. It will be apparent that various modifications may be made to the present invention without departing from the broader spirit and scope of the present disclosure set forth in the appended claims. Therefore, the present description and the accompanying drawings should be understood in an illustrative sense rather than a restrictive sense.
Claims
1. A method for operating an autonomous driving vehicle (ADV), the method comprising: generating a control command using a model predictive control (MPC) controller, the MPC controller comprising an optimizer using a cost function configured with one or more weights associated with a cost term of the cost function, wherein the control command is associated with a first trajectory of the ADV; applying the control command to a dynamic model of the ADV to simulate behavior of the ADV to generate a second trajectory of the dynamic model; adjusting the one or more weights based on an evaluation of the dynamics model in response to the control command to adjust the second trajectory of the dynamics model to track the first trajectory of the ADV, thereby resulting in an adjusted cost function for the MPC controller; generating, using the MPC controller with the adjusted cost function, a second control command for effecting movement of the ADV, and According to the second control command, the movement of the ADV is controlled.
2. The method according to claim 1, wherein: The evaluation of the dynamic model is performed relative to one or more current environmental conditions surrounding the ADV, the current environmental conditions including at least one of: pedestrians, other vehicles, road boundaries, or buildings, weather, wet or slippery roads, and traffic volume.
3. The method according to claim 2, wherein: Adjusting the one or more weights includes increasing a weight associated with a steering cost term of the one or more weights to constrain steering in response to the one or more current environmental conditions indicating that the road is wet or slippery.
4. The method according to claim 2, wherein: Adjusting the one or more weights includes reducing one of the one or more weights associated with a speed cost term to more aggressively increase speed in response to the one or more current environmental conditions indicating low traffic volume.
5. The method according to claim 2, wherein: Adjusting the one or more weights includes reducing one of the one or more weights associated with a braking cost term to increase braking in response to the one or more current environmental conditions indicating high traffic volume.
6. The method according to claim 1, wherein: The dynamic model of the ADV simulates the behavior of the ADV in a virtual environment, which includes a two-dimensional or three-dimensional representation of the current environment around the ADV.
7. The method according to claim 1, wherein: The dynamic model of the ADV includes one or more physical properties of the ADV, the physical properties including at least one of the following: braking condition, passengers or cargo in the ADV, battery charge state, or fuel level, wherein the one or more physical properties are representative of the current state of the ADV.
8. The method according to claim 1, wherein: The MPC controller includes a simplified model of the ADV for use with the cost function to generate an optimized control command sequence, the simplified model of the ADV having a less accurate representation of the ADV than a dynamic model of the ADV.
9. The method according to claim 1, wherein: Evaluation of the dynamic model in response to the control command is based on at least one of: proximity of the dynamic model to a pedestrian, another vehicle, or a building; speed of the dynamic model relative to a speed constraint; acceleration of the dynamic model relative to an acceleration constraint; position of the dynamic model relative to a road boundary or path; changes in travel direction or speed that could cause ride discomfort; and control force.
10. The method according to claim 1, wherein: The control command includes at least one of the following: a throttle command, a steering command, and a braking command.
11. The method according to claim 1, wherein: The cost term of the cost function includes at least one of: lateral trajectory error, driving direction error, speed cost, steering cost, acceleration cost, steering rate, braking and acceleration rate.
12. The method according to claim 1, wherein: Generating the control command includes: applying the MPC controller to a target control object to generate a control command sequence, and using a first control command in the control command sequence as the control command, wherein the control command sequence is optimized by minimizing the cost function while satisfying the target control object.
13. The method according to claim 12, wherein: Generating the second control command includes: applying the MPC controller with the adjusted cost function to the target control object to generate a second control command sequence, and using the first control command in the second control command sequence as the second control command, wherein the second control command sequence is optimized by minimizing the adjusted cost function while satisfying the target control object.
14. A non-transitory machine-readable medium storing instructions, which when executed by a processor causes the processor to perform operations of operating an autonomous driving vehicle (ADV), the operations comprising: generating a control command using a model predictive control (MPC) controller, the MPC controller comprising an optimizer using a cost function configured with one or more weights associated with a cost term of the cost function, wherein the control command is associated with a first trajectory of the ADV; applying the control command to a dynamic model of the ADV to simulate behavior of the ADV to generate a second trajectory of the dynamic model; adjusting the one or more weights based on an evaluation of the dynamics model in response to the control command to adjust the second trajectory of the dynamics model to track the first trajectory of the ADV, thereby resulting in an adjusted cost function for the MPC controller; generating, using the MPC controller with the adjusted cost function, a second control command for effecting movement of the ADV, and According to the second control command, the movement of the ADV is controlled.
15. The non-transitory machine-readable medium of claim 14, wherein: The evaluation of the dynamic model is performed relative to one or more current environmental conditions surrounding the ADV, the current environmental conditions including at least one of: pedestrians, other vehicles, road boundaries, or buildings, weather, wet or slippery roads, and traffic volume.
16. The non-transitory machine-readable medium of claim 15, wherein: Adjusting the one or more weights includes increasing a weight associated with a steering cost term of the one or more weights to constrain steering in response to the one or more current environmental conditions indicating that the road is wet or slippery.
17. A data processing system comprising: processor; as well as A memory coupled to the processor to store instructions that, when executed by the processor, cause the processor to perform operations for operating an autonomous driving vehicle ADV, the operations comprising: generating a control command using a model predictive control (MPC) controller, the MPC controller comprising an optimizer using a cost function configured with one or more weights associated with a cost term of the cost function, wherein the control command is associated with a first trajectory of the ADV; applying the control command to a dynamic model of the ADV to simulate behavior of the ADV to generate a second trajectory of the dynamic model; adjusting the one or more weights based on an evaluation of the dynamics model in response to the control command to adjust the second trajectory of the dynamics model to track the first trajectory of the ADV, thereby resulting in an adjusted cost function for the MPC controller; generating, using the MPC controller with the adjusted cost function, a second control command for effecting movement of the ADV, and According to the second control command, the movement of the ADV is controlled.
18. The data processing system according to claim 17, wherein: The evaluation of the dynamic model is performed relative to one or more current environmental conditions surrounding the ADV, the current environmental conditions including at least one of: pedestrians, other vehicles, road boundaries, or buildings, weather, wet or slippery roads, and traffic volume.
19. The data processing system according to claim 18, wherein: Adjusting the one or more weights includes increasing a weight associated with a steering cost term of the one or more weights to constrain steering in response to the one or more current environmental conditions indicating that the road is wet or slippery.
20. The data processing system according to claim 18, wherein: Adjusting the one or more weights includes reducing one of the one or more weights associated with a speed cost term to more aggressively increase speed in response to the one or more current environmental conditions indicating low traffic volume.
Citation Information
Patent Citations
Intelligent automobile multi-prediction-range model prediction trajectory tracking control method and system
CN110539752A
Unmanned vehicle path tracking and obstacle avoidance method
CN110703763A
