Probabilistic state tracking with multi-head measurement models

Abstract

A probabilistic system for tracking a vehicle state using non-synchronous collaboration of information includes a probabilistic multi-headed measurement model that associates incoming measurements with the vehicle state. A first head of the model associates measurements of satellite signals subject to measurement noise with a confidence in the vehicle state, and a second head associates estimates of the vehicle state subject to estimation noise with the confidence in the vehicle state. A probabilistic filter of the system recursively updates the confidence in the vehicle state based on the multi-headed measurement model that accepts one or a combination of the measurements of satellite signals subject to measurement noise and the estimates of the vehicle state subject to estimation noise.

Description

Technical Field

[0001] The present invention generally relates to positioning systems, such as the Global Positioning System (GPS) or the Quasi-Zenith Satellite System (QZSS), and more specifically, to using asynchronous cooperation of information to resolve vehicle status. Background Technology

[0002] A Global Navigation Satellite System (GNSS) is a satellite system used to determine the geographic location of a mobile receiver relative to Earth. Examples of GNSS systems include GPS, Galileo, GLONASS, QZSS, and BeiDou. Various GNSS correction systems are known, configured to receive GNSS signal data from GNSS satellites, process this GNSS data, calculate GNSS corrections from the GNSS data, and provide these corrections to the mobile receiver with the aim of achieving faster and more accurate calculation of the mobile receiver's geographic location.

[0003] Various position estimation methods are known, in which position calculation is based on repeated measurements of so-called pseudo-range and carrier phase observables by an Earth-based GNSS receiver. The "pseudo-range," or "code," observable represents the difference between the transmission time of a GNSS satellite signal and its local reception time, thus including the geometric distance covered by the satellite's radio signal. Measurements of the alignment between the carrier of the received GNSS satellite signal and a copy of this signal generated internally by the receiver provide another source of information for determining the line-of-sight between the satellite and the receiver. The corresponding observable is called "carrier phase," which represents the integral value of the Doppler frequency due to the relative motion between the transmitting satellite and the receiver.

[0004] Any pseudorange observation includes unavoidable error contributions, including receiver and transmitter clock errors, as well as additional delays caused by the non-zero refractive index of the atmosphere, instrument delay, multiplexing effects, and detector noise. Any carrier-phase observation also includes an unknown integer number of signal periods (i.e., an integer number of wavelengths) that have elapsed before a lock-on of the signal alignment has been achieved. This number is called the “carrier-phase ambiguity.” Typically, observable properties are measured, i.e., sampled by the receiver in discrete continuous time. The index of the time at which the observable property is measured is called an “epoch.” Known location determination methods typically involve dynamic numerical estimation and correction schemes based on distance and error components from measurements of observable properties sampled in consecutive epochs.

[0005] When GNSS signals are continuously tracked and no lock loss occurs, integer ambiguities resolved at the start of the tracking phase can be maintained throughout the entire GNSS positioning span. However, GNSS satellite signals may be occasionally obstructed (e.g., due to buildings in an "urban canyon" environment) or temporarily blocked (e.g., when the receiver passes under a bridge or through a tunnel). Typically, in such cases, integer ambiguity values ​​are lost and must be re-determined. This process can take anywhere from a few seconds to several minutes. In fact, significant multipath errors or unmodeled system biases in one or more pseudorange or carrier phase measurements can make ambiguity resolution difficult for existing commercial positioning systems. As receiver spacing (i.e., the distance between the reference receiver and the moving receiver whose position is being determined) increases, distance-dependent biases (e.g., orbital errors and ionospheric and tropospheric effects) also increase, making reliable ambiguity resolution (or re-initialization) an even greater challenge. Furthermore, lock loss, known as cycle slips, can also occur due to interruptions in the receiver's continuous phase lock on the signal. Cycle slips can be caused, for example, by power loss, receiver software malfunctions, or faulty satellite oscillators. Furthermore, altering ionospheric conditions can also lead to cycle slips.

[0006] GNSS enhancement refers to techniques used to improve the accuracy of positioning information provided by the Global Positioning System or other Global Navigation Satellite Systems (typically satellite networks used for navigation). For example, some methods use differential techniques based on inter-satellite differentials, inter-receiver differentials, inter-epoch differentials, and combinations thereof. Single and double differentials between satellites and receivers reduce error sources, but do not eliminate them.

[0007] Therefore, there is a need to improve the accuracy of GNSS positioning. To address this issue, many different methods utilize the cooperation of multiple GNSS receivers to enhance GNSS positioning accuracy. However, for proper cooperation to occur, the multiple GNSS receivers need to be synchronized, and their operation needs to be constrained. For example, U.S. Patent 9,476,990 describes cooperative GNSS positioning estimation via multiple mechanically linked modules. However, such constraints on cooperation to enhance the accuracy of GNSS positioning are not always practical. Summary of the Invention

[0008] Some implementations are based on the understanding that current methods for tracking vehicle state based on satellite signals received from a Global Navigation Satellite System (GNSS) assume either individual or centralized estimation based on internal vehicle modules, or distributed estimation performed in a tightly controlled and / or synchronized manner. Examples of such distributed estimation include decentralized systems that determine different aspects of state tracking and estimate vehicle state through consensus; unbalanced systems that track the state of the system independently while one type of tracking dominates over another; and distributed systems that include multiple synchronized GNSS receivers, preferably located at fixed distances from each other.

[0009] Some implementations recognize the advantages of cooperative state tracking when the vehicle's internal modules use additional information determined externally. However, some implementations are based on the understanding that such external information is not always available. Therefore, when tracking is performed by the vehicle's internal modules, cooperative but asynchronous state tracking is required, but when such information is available, external information can be seamlessly integrated.

[0010] Some implementations of various probabilistic filters that recursively track vehicle states consist of a two-part implementation. The first part estimates samples of vehicle states, and the second part updates the probability distribution of the states based on these estimated samples. At first glance, these two parts seem integrated, as the execution of one part is inseparable from the other. However, some implementations are based on the understanding that these two parts are related through causal dependence rather than temporal dependence. Specifically, the probability update depends on the existence of state sample estimates; in this sense, the arrival of a new state sample should trigger an update. However, when and how the state samples are estimated is irrelevant to the update of the probability distribution.

[0011] Some implementations utilize this understanding to disambiguate the probabilistic state estimation and update portions. This disambiguation allows the reuse of probabilistic updates regardless of when and how the state samples were estimated. Thus, state estimation can be determined by an internal vehicle module based on GNSS measurements, or by an external module if available. Regardless of the principle behind state estimation, probabilistic updates operate in the same manner, and their quality depends not on the frequency of new information arrivals, but on the quality of the estimation itself. In this way, internal and external state estimators can operate asynchronously.

[0012] Based on these understandings, some implementations disclose a multi-head measurement model that allows probabilistic filters to update the confidence level of a vehicle state using different sources. The multi-head measurement model includes multiple paths that produce different types of outputs but have the same probabilistic structure. For example, in some implementations, the multi-head measurement model includes two heads. A first head correlates the measurement results of the satellite signal with the confidence level of the vehicle state subjected to measurement noise, and a second head correlates the estimate of the vehicle state with the confidence level of the vehicle state subjected to estimated noise. Therefore, the measurement model includes different information and different types of noise, but has a similar structure acceptable to probabilistic filters, thus allowing the outputs of different heads to be used individually or in combination.

[0013] In this way, in some implementations, the probabilistic filter recursively propagates the parameters of the probability distribution of the vehicle state according to the motion model of the vehicle's state transitions under processed noise at each update step, and updates the parameters of the probability distribution upon receiving the output of one or a combination of the first head and the second head of the multi-head measurement model.

[0014] For example, the first head allows updating the confidence level of the vehicle state using GNSS measurements with specific uncertainties, biases, and ambiguities that affect the accuracy and / or quality of the state estimation. The output of the first head is typically available at the same update frequency as the probabilistic filter. Conversely, the first head allows updating the confidence level of the vehicle state using external information, which may only be received occasionally but can have higher state estimation accuracy. For example, the second head can process state estimates received from an external measurement module. Therefore, the second head may receive information less frequently, but regardless of the execution frequency of the first and / or second heads in the multi-head measurement model, the outputs of the first and second heads are seamlessly integrated without violating the probabilistic guarantees of the probabilistic filter.

[0015] Some implementations are based on the understanding that the external measurement module can be passive, active, or a combination of both. In passive mode, the external measurement module receives an estimate of the vehicle state independently of GNSS measurements performed by the internal measurement module. For example, the vehicle may pass near a roadside unit (RSU) configured to estimate its state using various telemetry techniques. In active mode, the vehicle may send current information about its vehicle state to a remote server (which could also be the RSU) and, in response, receive external measurement results corrected based on internal measurement results. Here, the current information includes the mean of the vehicle state estimate, the covariance of the vehicle state estimate, a set of codes used by the internal measurement module to update its internal state, and carrier phase measurements.

[0016] Therefore, some implementations aim to provide a cooperative but asynchronous estimation of vehicle states used by probabilistic state estimation. Another objective of some implementations is to provide such cooperative asynchronous estimation when an external active module performs an update at a rate that may differ from that performed by the internal probabilistic filter.

[0017] Some implementations are based on the understanding that the GNSS positioning problem involves estimating the state of a receiver from a set of codes and carrier phase measurements received from one or more satellite constellations. The measurement equations involved are time-varying, non-linear in their application to the receiver's position, and incorporate various biases and integer ambiguities. Within the carrier phase measurements, there exist integer biases known as ambiguities, unique to each carrier phase measurement for each satellite. When these time biases are taken into account, they follow integer jumps, remaining constant until they occasionally and independently jump to new integer values, often referred to as "cycle slips." Therefore, the GNSS positioning problem can be viewed as a mixed-integer GNSS positioning problem, solving a mixed-integer problem of ever-increasing size (each time step containing new integer biases), and requiring careful consideration of how best to relax the estimation problem to enable the final algorithm to be implemented.

[0018] Some implementations are based on the understanding that, in the setting of the probabilistic filter, the uncertainty of integer ambiguity caused by, for example, cycle slips is reflected in the second moment (i.e., covariance) of the probability distribution of the vehicle state. To this end, in order to benefit from the active mode of the external measurement module, some implementations send parameters of the probability distribution to allow a remote server to correct this information. The remote server can collect similar information from other nearby vehicles, fuse the received information to correct the parameters of the probability distribution of the vehicle state, and send the updated probability parameters back to the vehicle. Upon receiving the updated probability parameters, the external measurement module estimates the vehicle state at the current moment consistent with the updated parameters and triggers the execution of the probabilistic filter. For example, the external measurement module can sample the updated probability distribution defined by the propagated mean and the received variance based on the mean of the probability distribution propagated over time according to the vehicle's motion model, and update the probability distribution of the vehicle state using a probabilistic filter with sampled state estimates.

[0019] In some implementations, the vehicle's state includes integer ambiguities related to carrier phase measurements from a specific satellite, the vehicle's position, its speed, and residual states modeled with atmospheric (e.g., ionospheric and tropospheric) delays. For example, some methods use differential techniques based on inter-satellite, inter-receiver, inter-epoch, and combinations thereof. Single and double differences between satellites and receivers reduce error sources but do not eliminate them, thus reducing the accuracy of state estimation.

[0020] Some implementations are based on the understanding that ignoring state biases (such as ionospheric errors) can lead to slight inaccuracies in state estimation. This is because biases are typically eliminated by single or double differences in GNSS measurements. This solution works well when the desired accuracy of vehicle position estimation is on the order of meters, but can be problematic when the desired accuracy is on the order of centimeters. To address this, some implementations include state biases in the vehicle state and identify them as part of the state tracking provided by probabilistic filters.

[0021] Some implementations are based on the understanding that when the external state estimate determined at the remote server is sufficiently accurate, a multi-head measurement model that can communicate asynchronously but cooperatively with the external server can solve the problem of state estimation accuracy. However, the server still needs to determine the state with the desired accuracy.

[0022] Some implementations recognize that accurately determining the atmospheric time delay for a single vehicle may be impractical because, for a single vehicle, probabilistic filters would need to address both integer ambiguity and uncertainty in the atmospheric delay. However, for vehicles that are sufficiently close (e.g., 2-10 km), the atmospheric delays are similar to each other. Therefore, for an external active module that collects similar information from multiple vehicles, there is significant overlap between the states of different vehicles, allowing the external measurement module to update the state of a single vehicle in ways that a probabilistic filter for a single vehicle cannot achieve. For example, ionospheric state deviations can be shared among vehicles that are close to each other.

[0023] Additionally or alternatively, some implementations are based on the understanding that the inaccuracy of the state estimate determined by the probabilistic filter may be caused by ambiguity and residual delay, or by measurement noise. Some implementations are based on the understanding that when the states of multiple vehicles are cooperatively determined, the vehicle states may or may not overlap, and ambiguity or delay may or may not overlap, but the measurement noise of different vehicles has correlations that can be explored through cooperative state estimation. For example, errors in the measurement results of different pairs of vehicles may be correlated.

[0024] To this end, some implementations perform probabilistic tracking of multiple vehicles based on augmentation probabilistic filters that combine information from different vehicles into the augmentation domain. For example, the augmentation probabilistic filter fuses the states of multiple vehicles into an augmented state, and fuses the measurements of satellite signals from multiple vehicles into an augmented measurement result subjected to augmented measurement noise defined by an off-diagonal covariance matrix. This off-diagonal covariance matrix has non-zero off-diagonal elements that capture the correlation between the measurement noise of different vehicles. This correlation is additional information that the augmentation probabilistic filter can explore for jointly tracking the states of multiple vehicles.

[0025] In some implementations, the probabilistic filter is a mixed-integer least-squares Kalman filter, where the nonlinear measurement equation is linearized around its current estimate, resulting in a mixed-integer extended Kalman filter. Other implementations recognize that the linearization in the EKF can be inaccurate. Therefore, in some implementations, the probabilistic filter is a mixed-integer linear regression Kalman filter. Such a filter more accurately represents the first and second moments of the probability distribution, but is significantly more computationally expensive.

[0026] In one implementation, it can be recognized that the vehicle state portion (e.g., integer ambiguity) is linear in the model, but the position is non-linear. Therefore, portions of the first and second moments can be determined analytically, while the remaining portions are estimated using a linear regression Kalman filter.

[0027] The GNSS positioning problem can be viewed as a mixed-integer GNSS positioning problem, which addresses a mixed-integer problem with ever-increasing dimensions (each time step contains a new integer bias). Careful consideration is needed on how best to relax this estimation problem to make the resulting algorithm feasible. In particular, for external estimation of the state when combining information from multiple vehicles, the resulting estimation problem becomes very high-dimensional, thus computationally expensive.

[0028] In some implementations, relaxation is addressed by determining two related but independent probability distributions, one of which considers only real-valued parameters, i.e., ignoring that the ambiguity is actually an integer, while the second distribution is determined based on integer ambiguities determined from the first distribution by solving an optimization problem.

[0029] Therefore, one embodiment discloses a probabilistic system for tracking vehicle state using asynchronous cooperation between information received from satellite signals transmitted by a Global Navigation Satellite System (GNSS) and information transmitted via a radio frequency (RF) channel. The probabilistic system includes a memory configured to store a probabilistic multi-head measurement model that associates incoming measurement results with vehicle state, wherein the probabilistic multi-head measurement model includes a first head that associates measurement results of satellite signals subjected to measurement noise with a confidence level of the vehicle state, and a second head that associates an estimate of the vehicle state subjected to estimation noise with a confidence level of the vehicle state; and at least one processor configured to process executable instructions for modules of the probabilistic system.

[0030] The probabilistic system modules include: a GNSS measurement module operatively connected to a GNSS receiver and configured to determine the measurement results of satellite signals and measurement noise; an RF measurement module operatively connected to the vehicle's RF receiver and configured to receive data indicating an estimate of the vehicle state and estimation noise; and a probabilistic filter configured to recursively update parameters of the probability distribution of the vehicle state to generate a confidence level of the vehicle state based on a multi-head measurement model that accepts one or a combination of the measurement results of satellite signals subjected to measurement noise and the estimate of the vehicle state subjected to estimation noise. Attached Figure Description

[0031] [ Figure 1A ]

[0032] Figure 1A A schematic diagram of a Global Navigation Satellite System (GNSS) according to some implementations is shown.

[0033] [ Figure 1B ]

[0034] Figure 1B An example is shown when two other vehicles, 140 and 150, equipped with GNSS receivers are present.

[0035] [ Figure 1C ]

[0036] Figure 1C Various variables, used individually or in combination, in modeling motion and / or measurement models according to some implementations are shown.

[0037] [ Figure 1D ]

[0038] Figure 1D A schematic diagram is shown of a scenario where multipath interference affects the receiver's signal.

[0039] [ Figure 1E ]

[0040] Figure 1E A conceptual diagram of a probabilistic filter used in some implementations is shown.

[0041] [ Figure 1F ]

[0042] Figure 1F A schematic diagram of asynchronous reception of different signals according to some implementation methods is shown.

[0043] [ Figure 1G ]

[0044] Figure 1G A schematic diagram of the structure of a multi-head measurement model according to some embodiments is shown.

[0045] [ Figure 1H ]

[0046] Figure 1H A flowchart illustrating an example of how to update the first and second moments of the state estimate at a time instance, according to one implementation, is shown.

[0047] [ Figure 1I ]

[0048] Figure 1I Another flowchart of a method for updating the first and second moments of a state estimate according to another embodiment is shown.

[0049] [ Figure 1J ]

[0050] Figure 1J The opposite case is shown, in which, according to another embodiment, only the first and second moments of the state estimation are received at a specific time instance.

[0051] [ Figure 1K ]

[0052] Figure 1K A flowchart illustrating an example of the process is shown.

[0053] [ Figure 2A ]

[0054] Figure 2A A flowchart is shown of a method for tracking vehicle status based on asynchronous collaboration of usage information according to some implementations.

[0055] [ Figure 2B ]

[0056] Figure 2B A probabilistic system for tracking vehicle status based on asynchronous cooperation of usage information in some implementations is shown.

[0057] [ Figure 2C ]

[0058] Figure 2C A flowchart illustrating an example of a method for determining an estimate of 240a according to some embodiments is shown.

[0059] [ Figure 2D ]

[0060] Figure 2D Examples of methods for determining confidence levels regarding vehicle status, according to some implementations, are shown.

[0061] [ Figure 3A ]

[0062] Figure 3A A flowchart is shown of an example implementation of a method for vehicle state estimation according to some embodiments.

[0063] [ Figure 3B ]

[0064] Figure 3B A flowchart of a method for determining carrier phase ambiguity according to some embodiments is shown.

[0065] [ Figure 3C ]

[0066] Figure 3C A flowchart is shown as an example implementation of a method for sampling floating-point values ​​consistent with a motion model and its processing noise, according to one embodiment.

[0067] [ Figure 3D ]

[0068] Figure 3D A schematic diagram is shown illustrating some of the principles employed by some implementation methods.

[0069] [ Figure 4 ]

[0070] Figure 4 A flowchart is shown of a method for executing a probabilistic filter to update the parameters of a probability distribution of a vehicle state, according to one embodiment.

[0071] [ Figure 5A ]

[0072] Figure 5A A traffic scenario is shown according to some implementation methods.

[0073] [ Figure 5B ]

[0074] Figure 5BThe exchange of synchronization messages in synchronization based on the IEEE Precision Time Protocol (PTP) is illustrated.

[0075] [ Figure 5C ]

[0076] Figure 5C A probabilistic system for tracking vehicle status based on asynchronous cooperation of usage information in some implementations is shown.

[0077] [ Figure 5D ]

[0078] Figure 5D A traffic scenario according to another implementation is shown.

[0079] [ Figure 6A ]

[0080] Figure 6A A flowchart is shown of a recursive method for determining a first moment and a second moment of a vehicle state using multiple transmissions from multiple vehicles, according to some embodiments.

[0081] [ Figure 6B ]

[0082] Figure 6B A flowchart is shown of a recursive method for updating parameters of an enhanced state at a remote server, according to some implementations.

[0083] [ Figure 6C ]

[0084] Figure 6C A flowchart is shown of a method for fusing multiple states of a vehicle with an enhanced state according to some embodiments.

[0085] [ Figure 6D ]

[0086] Figure 6D An example of overlapping state spaces according to some implementations is shown.

[0087] [ Figure 6E ]

[0088] Figure 6E A flowchart of a method for fusing enhanced state with vehicle state according to some embodiments is shown.

[0089] [ Figure 6F ]

[0090] Figure 6F A flowchart of a method for determining cross-covariance according to one embodiment is shown.

[0091] [ Figure 6G]

[0092] Figure 6G An example of cross-covariance in measurement noise of a full measurement model comprising measurement results from multiple transmissions, according to some implementations, is shown.

[0093] [ Figure 6H ]

[0094] Figure 6H A method for determining the measurement noise matrix is ​​shown to generate Figure 6G The flowchart of the structure method in the document.

[0095] [ Figure 7 ]

[0096] Figure 7 A probabilistic system is shown, according to some embodiments, for tracking the state of multiple vehicles by asynchronous cooperation of information received from vehicles and transmitted via an RF channel.

[0097] [ Figure 8 ]

[0098] Figure 8 An example of state estimation-based vehicle-to-vehicle (V2V) communication and planning according to one implementation is shown.

[0099] [ Figure 9 ]

[0100] Figure 9 This is a schematic diagram of multi-vehicle queuing formation for an accident avoidance scenario according to one embodiment.

[0101] [ Figure 10 ]

[0102] Figure 10 A block diagram of a system 1000 for direct and indirect control of a hybrid autonomous vehicle, according to some embodiments, is shown.

[0103] [ Figure 11A ]

[0104] Figure 11A A schematic diagram of a vehicle 1101 directly or indirectly controlled according to some embodiments is shown.

[0105] [ Figure 11B ]

[0106] Figure 11B A schematic diagram of the interaction between a controller 1102 that receives controlled commands from a system 1000 and a controller 1100 of a vehicle 1101, according to some embodiments, is shown. Detailed Implementation

[0107] Figure 1AA schematic diagram of a Global Navigation Satellite System (GNSS) according to some embodiments is shown. For example, the Nth satellite 102 transmits codes 120 and 121 and carrier phase measurements to a set of receivers 130 and 131. For example, receiver 130 is positioned to receive signals 110 and 120 from N satellites 101, 103, 104, and 102. Similarly, receiver 131 is positioned to receive signals 121 and 111 from N satellites 101, 103, 104, and 102.

[0108] In various implementations, GNSS receivers 130 and 131 can be of different types. For example, in Figure 1A In some exemplary embodiments, receiver 131 is a reference receiver whose location is known. For example, receiver 131 may be a receiver mounted on the ground. Conversely, receiver 130 is a mobile receiver configured to move. For example, receiver 130 may be mounted in a mobile phone, vehicle, or tablet. In some implementations, a second receiver 131 is optional and may be used to remove or at least reduce uncertainties and errors caused by various sources, such as atmospheric effects and errors in the internal clocks of the receiver and the satellite. In some implementations, multiple GNSS receivers are present to receive the code and carrier phase signal.

[0109] Figure 1B An example is shown when two additional vehicles 140 and 150 equipped with GNSS receivers are present. The satellite can transmit code and carrier phase signals to multiple receivers, allowing the receivers to share the same source of code and carrier phase signals. For example, satellite 104 can transmit signals to both receiver 140 and receiver 150, and satellite 101 can transmit signals to all three receivers 130, 140, and 150. Additionally or alternatively, receivers of the GNSS signal can exchange information with each other on radio waves 161.

[0110] Some embodiments aim to disclose a system and method for improving satellite-based tracking of vehicle status, wherein the vehicle is equipped with a GNSS receiver. Another objective is to provide a system and method that utilizes asynchronous cooperation of information received from satellite signals. Yet another objective of some embodiments is to provide a system and method that is probabilistic, i.e., it takes into account probabilistic perturbations and error sources. Other embodiments aim to use different information from different sources to track vehicle status, rather than relying solely on satellite signals. For example, in some embodiments, vehicle status is tracked using a first moment and a second moment of the probability distribution of the vehicle status received from a satellite using a GNSS receiver and from a remote server using a radio frequency (RF) receiver.

[0111] Alternatively or concurrently, some implementations aim to provide a filter that is probabilistic. This filter is a probabilistic filter, such as a state estimation filter, which provides state estimates based on motion and measurement models.

[0112] An example of a state estimator is the Kalman filter, which uses a series of measurements observed over time, including statistical noise and other inaccuracies, and produces estimates of unknown variables by estimating the joint probability distribution of the variables across each time frame. These estimates tend to be more accurate than estimates based solely on a single measurement. The Kalman filter keeps track of the estimated system state and the uncertainty of the estimate. The estimate is updated using a motion model of the state transitions and the measurements. Some implementations use a Kalman filter-based system with a motion model subjected to processing noise from a GNSS receiver and a measurement model of satellite signals subjected to measurement noise. In some implementations, the measurement model is probabilistic and multi-headed, i.e., it includes multiple paths that produce different types of outputs. In some implementations, the multi-headed measurement model includes two heads. The first head correlates the measurement of the satellite signal with a confidence level of the state of the vehicle subjected to measurement noise, and the second head correlates the estimate of the vehicle state with a confidence level of the state of the vehicle subjected to estimated noise. Thus, the measurement model includes different information and different types of noise, but has a similar structure acceptable to probabilistic filters, allowing the outputs of different heads to be used individually or in combination.

[0113] In this way, in some implementations, the probabilistic filter recursively propagates the parameters of the probability distribution of the vehicle state according to the motion model of the vehicle's state transitions under processed noise at each update step, and updates the parameters of the probability distribution upon receiving the output of one or a combination of the first head and the second head of the multi-head measurement model.

[0114] In some implementations, the motion model of the receiver is one with state vectors. A general kinematic constant acceleration model is used, where the three components are the receiver's position, velocity, and acceleration. In some other implementations, the temporal evolution ambiguity of the satellite signal propagation is modeled as... ,in It's about ambiguity, and It has covariance Gaussian noise processing.

[0115] In some implementations, ambiguity is incorporated into the vehicle's state. Other implementations include a bias state that captures residuals in atmospheric delays (e.g., ionospheric delays). For receivers that are sufficiently close to each other, the ionospheric delays of different vehicles are the same or very similar. Some implementations utilize this relationship to resolve these delays and / or ambiguities.

[0116] Some implementations capture measurement models The carrier and code signals in the signal, where, is the measurement noise, h is the nonlinear part of the measurement equation depending on the receiver's location, n is the integer ambiguity, λ is the wavelength of the carrier signal, and y is the single or double difference between the satellite combinations K.

[0117] In some implementations, probabilistic filters use carrier phase single difference (SD) and / or double difference (DD) to estimate the receiver state indicating the receiver's location. When a carrier signal transmitted from one satellite is received by two receivers, the difference between the first carrier phase and the second carrier phase is called the single difference (SD) in the carrier phase. Alternatively, SD can be defined as the difference between signals from two different satellites arriving at the receiver. For example, when the first satellite is referred to as the reference satellite, the difference may come from both the first and second satellites. For example, the difference between signal 110 from satellite 101 and signal 120 from satellite 102 is an SD signal, where satellite 101 is the reference satellite. Figure 1A For receivers 131 and 130, the difference between the SD and SD in the carrier phase obtained from the radio signals of the two satellites is called the double difference (DD) in the carrier phase. This carrier phase difference is then converted to wavelengths, for example, for L1 GPS (and / or GNSS) signals. When numbering, the number is separated into a decimal part and an integer part. The decimal part can be measured by a positioning device, but the integer part cannot be directly measured by the positioning device. Therefore, the integer part is called the integer deviation or integer ambiguity.

[0118] Typically, GNSS can use multiple constellations simultaneously to determine receiver status. For example, GPS, Galileo, GLONASS, and QZSS can be used concurrently. Satellite systems typically transmit information on up to three different frequency bands, and for each band, each satellite transmits code measurements and carrier phase measurements. These measurements can be combined into single-difference or double-difference methods, where single-difference involves taking the difference between a reference satellite and other satellites, and double-difference involves taking the difference between the receiver of interest and a reference receiver with a known static position.

[0119] Figure 1C Various variables, used individually or in combination, are illustrated in the modeling of motion and / or measurement models according to some embodiments. Some embodiments utilize measurement models to model the carrier and code signals at each frequency.

[0120] in, This is the result of code measurement. The distance between the receiver and the j-th satellite is denoted by c, where c is the speed of light. It's the receiver clock skew. It's a satellite clock skew. It is ionospheric delay. It is a tropospheric delay. It is probabilistic code observation noise. This is the carrier phase observation result, where λ is the carrier wavelength. It is an integer ambiguity, and It is probabilistic carrier observation noise.

[0121] In one implementation, most error sources can be eliminated by transforming the original measurement model using a reference receiver b installed at a known location to the original receiver r. For example, one implementation would... Figure 1A The difference between the two receivers 130 and 131 in the middle is formed as and This eliminates errors caused by satellite clock skew. Another implementation involves creating a double difference between the two satellites, j and l. By doing so, clock error terms caused by the receivers can be removed. Furthermore, for short distances between the two receivers (e.g., 30 km), ionospheric errors can be ignored, at least where centimeter accuracy is not required, resulting in... , Alternatively, one implementation creates a difference between two satellites 101 and 102, resulting in SD measurement results.

[0122] Alternatively or concurrently, some implementations are based on the understanding that ignoring state biases (such as ionospheric errors) can lead to slight inaccuracies in state estimation. This is because biases are typically removed through single or double differences in GNSS measurements. This solution works well when the desired accuracy of vehicle position estimation is on the order of meters, but can be problematic when the desired accuracy is on the order of centimeters. To address this, some implementations include state biases in the vehicle state and identify these state biases as part of the state tracking provided by probabilistic filters.

[0123] In certain scenarios, such as deep urban canyons, satellite signals suffer from multiple distortions, making it difficult to perform high-precision state estimation due to the information contained in the code and carrier phase signals. For example, Figure 1DA schematic diagram illustrating the signal situation of multipath interference receiver 101d is shown. Receiver 101d receives various signals 109d and 119d from satellites 110d and 120d. Other satellites 130d and 140d also transmit signals 128d, 129d, 138d, and 139d, but these signals are not directly transmitted to the receiver due to obstacles 170d (e.g., buildings in urban areas).

[0124] Previously, signal 138d transmitted from satellite 140d was unavailable, but suddenly, satellite signal 139d arrives at the receiver after multipath 102d. This situation can severely impair the performance of the probabilistic filter in tracking vehicle states because the estimator is locked onto an incorrect ambiguity estimate, resulting in a large estimation error. However, there may be other receivers nearby without such problems; for example, a receiver may have multipath and few visible satellites due to obstruction by large buildings, while vehicles several blocks away have good estimation performance. In some implementations, this is utilized by updating the parameters of the probability distribution in the probabilistic filter using measurements from the GNSS measurement module and a long-range estimate of the vehicle state, where the long-range estimate is determined using data from multiple vehicles.

[0125] An example of a probabilistic filter is the Kalman filter. Figure 1E A schematic diagram of a Kalman filter (KF) used for state estimation by several implementations is shown. The KF is a tool for state estimation in linear state-space models, and it is the optimal estimator when the noise source is known and is Gaussian, in which case the state estimate is Gaussian distributed. The KF estimates the mean and variance of the Gaussian distribution because the mean and variance are sufficient statistics for describing the Gaussian distribution.

[0126] KF begins with initial knowledge of the state 110e to determine the mean and variance of the state 111e. KF then uses a model of the system (e.g., a motion model of the vehicle) to predict 120e the state and variance to the next time step, obtaining the updated mean and variance of the state 121e. KF then uses the measurement results 130e in the update step 140e using the system's measurement model to determine the updated mean and variance of the state 141e. The output 150e is then obtained, and the process is repeated in the next time step 160e.

[0127] Some implementations employ probabilistic filters including various variants of KF, such as extended KF (EKF), linear-regression KF (LRKF), and unscented KF (UKF). Despite the existence of multiple variants of KF, they are conceptually similar... Figure 1E It works as shown. It is worth noting that KF uses the measurement results 130e described by the probabilistic measurement model to update the first and second moments of the probability distribution of interest, namely, the mean and covariance. In some implementations, the probabilistic measurement model is a multi-head measurement model constructed to satisfy the principle of measurement update in KF.

[0128] Some implementations are based on the understanding that the outputs of different heads in a multi-head measurement model do not need to be synchronized, thus allowing signals to be received at different time steps. Other implementations are based on the understanding that, in order to perform KF as is, the measurement model needs to be constructed such that KF can accept the measurement model.

[0129] Figure 1F A schematic diagram illustrating asynchronous reception of different signals according to some implementations is shown. GNSS signals are received at certain time steps over time, and a first moment and a second moment are received at other time steps. For example, a GNSS signal, along with the first and second moments, is received at time 110f. However, a second GNSS signal is received at time 120f, while the first and second moments are received at time 130f. That is, sometimes the signals are received simultaneously, and sometimes not. To handle such signals that may be received at different time steps, the measurement model is constructed in a form that allows different signals to be used independently of each other in the KF.

[0130] Therefore, in some implementations, the measurement model is probabilistic and multi-headed, i.e., it includes several parts, wherein a first head correlates the measurement results of the satellite signal with a confidence level of the state of the vehicle subjected to measurement noise, and wherein a second head correlates the estimate of the vehicle state with a confidence level of the state of the vehicle subjected to estimated noise. Thus, the measurement model incorporates different information and different types of noise.

[0131] In one implementation, a first head of the measurement model subjected to measurement noise is used to correlate GNSS measurements with confidence levels of vehicle state, while a second head of the measurement model subjected to measurement noise is used to correlate estimates of vehicle state with confidence levels of vehicle state received through other means.

[0132] Some implementations are based on the principle of using a measurement model to process measurement results in order to correct and update the first and second moments of 140e using the measurement results 130e and the measurement model. In order to use KF, the measurement model and measurement noise must be written in probabilistic form so that they can be processed in an equalized manner, ensuring that the measurement results are not incorrectly scaled proportionally to the first and second moments.

[0133] Other implementations are based on the understanding that the probabilistic structure needs to be coordinated between the first and second heads of the measurement model. Specifically, regardless of whether the update step 140e of the probability distribution of the vehicle state is performed using GNSS signals or the received vehicle state estimates and estimated noise (e.g., represented as the first and second moments), the output has the same structure.

[0134] Figure 1G A schematic diagram of the structure of a multi-head measurement model 100g according to some embodiments is shown. The multi-head measurement model accepts input, uses a first head 110g of the measurement model to correlate GNSS measurement results with a confidence level of the vehicle state, and uses a second head 120g to correlate an alternative estimate of the vehicle state with the confidence level of the vehicle state. For example, the multi-head measurement model accepts GNSS measurement results 130g and measurement noise 140g. The multi-head measurement model then transforms the measurement noise 140g into a second moment representing the measurement noise 150g. For example, a moment with zero mean and covariance is used. The measurement noise 150g is represented by a Gaussian distribution. Using the representation 150g of the measurement noise, the first head 110g correlates the measurement result 130g with the confidence level 160g of the vehicle state 111g. The output 112g from the first head of the measurement model is in a form acceptable to KF. For example, in one embodiment, the output 112g is a relation 111g that describes the error between the measurement result and confidence level of the state inserted into the measurement model and the measurement noise. In another embodiment, the output 112g is a relation 111g decomposed into the measurement result, the confidence level of the state inserted into the measurement model, and the measurement noise covariance.

[0135] In one implementation, the input to the second head of the measurement model is an estimate of the vehicle state 170g and an estimated noise 180g. The multi-head measurement model converts the estimated noise 190g into the same format as the conversion 150g. For example, if the measurement noise is described by a Gaussian distribution, then the estimated noise needs to be described by a Gaussian distribution. Using the state estimate and the representation of the estimated noise 190g, the second head 120g correlates the estimate 121g with the confidence level 160g of the vehicle state. In some implementations, the correlation is an affine relation. ,in, It is an estimated noise expressed in a specific form. It is the confidence level of the vehicle state, for example, the first moment determined by a probabilistic filter or a sample representation of the distribution determined by a probabilistic filter.

[0136] Some implementations are based on the understanding that the representation of outputs 112g and 122g in the same probabilistic form enables updates in KF to be performed according to the same principle, regardless of whether the updated information is 112g or 122g, i.e., whether the information comes from the first head or the second head.

[0137] Other implementations are based on the understanding that updates using the first header and updates using the second header can be performed sequentially or jointly. Figure 1H A flowchart illustrating an example of how to update the first and second moments at a time instance according to one implementation is shown. Using the output 112g from a multi-head measurement model, the method updates the first and second moments at 110h using the first head of the measurement model. Then, using the output 122g from the second head, the updated moment 111h at 120h is updated again, resulting in the updated moment 121h.

[0138] Other implementations are based on the understanding that GNSS measurements and the received vehicle state estimates, as well as the estimated noise, may not be received at the same time instance, such as... Figure 1F As shown. Figure 1I Another flowchart is shown, for example, of a method for updating the first and second moments using KF. Using the output 112g from the first head of the measurement model, this method updates moment 110h and moment 111h.

[0139] Figure 1J A contrasting scenario is illustrated, where only the first and second moments are received at a specific time instance. Therefore, one implementation uses the output 122g from the second head of the measurement model to update the 110h moment 111j.

[0140] Other implementations are based on the understanding that updates using the first and second headers can be performed in conjunction, i.e., simultaneously. Figure 1K A flowchart illustrating an example of such a process is shown. Using the output 112g from the first head and the output 122g from the second head, the method updates moment 110k using both the first and second heads, resulting in the updated moment 111k. In one implementation, joint updates are performed by stacking the two heads of the measurement model, thereby producing an enhanced output of the measurement model. That is, GNSS measurement results and estimates of vehicle state, along with their correlations, are combined to produce a large joint update.

[0141] Figure 2AA flowchart illustrating a method for tracking vehicle status using asynchronous cooperation of usage information according to some embodiments is shown. In the embodiments, information is received from satellite signals transmitted from a Global Navigation Satellite System (GNSS) and also transmitted via a radio frequency (RF) channel. First, according to some embodiments, the method receives 210a signals 210a transmitted from a set of satellites 101, 102, 103, 104 from a GNSS receiver 209a. These signals include code signals and carrier phase signals, wherein each carrier signal includes carrier phase ambiguity as an unknown integer wavelength of the carrier signal traveling between satellite 101, 102, 103, or 104 and vehicle 130. Using GNSS signal 215a, the method determines 220a the measurement results of the satellite signals and the corresponding measurement noise 225a. The method also receives 230a an estimate of the vehicle status and an estimated noise determined by an external server from the vehicle's RF receiver 229a via RF signal 235a. In various embodiments, the estimate and estimated noise 235a are probabilistically determined as a first moment and a second moment 235a of the vehicle status. Alternatively or alternatively, various probabilistic techniques can be used to convert the estimated noise 235a into the first moment and the second moment of the vehicle state.

[0142] Using a first moment and a second moment, the method determines an estimate of the vehicle state 240a and the associated estimation noise of the vehicle state estimate. Using the estimate 245a, the measurement result 225a, and the multi-head measurement model 249a, the method uses the first head of the measurement model subjected to measurement noise to correlate the GNSS measurement result 225a with the confidence level of the vehicle state 250a, and uses the second head of the measurement model to correlate the state estimate with the confidence level of the vehicle state subjected to estimation noise 250a. Using the associated measurement result and estimate 255a, the method recursively updates the parameters of the probability distribution of the vehicle state 260a to produce a confidence level 265a regarding the vehicle state.

[0143] Figure 2BA probabilistic system 200 according to some embodiments is illustrated, which is used to track vehicle status using asynchronous cooperation between information received from satellite signals 209 transmitted from a navigation satellite system (GNSS) and information transmitted via a radio frequency (RF) channel 239. The probabilistic system includes a GNSS receiver 220 for receiving the satellite signals 209. In one embodiment, the signals include code signals, carrier phase signals, timestamps, navigation messages, and observation messages required to determine various delays and interferences affecting the satellite signals. For example, the navigation message may include parameters required to determine atmospheric delay. For example, if the Klobuchar atmospheric delay model is used, the navigation message includes parameters for determining ionospheric time delay based on the ionosphere's effect on signal propagation determined by the atmospheric delay model. In other embodiments, such a message may include parameters for determining ionospheric delay based on an SBAS delay model.

[0144] The system includes a memory 280 for storing a probabilistic multi-head measurement model 281 that correlates input measurement results with vehicle state. In some embodiments, the multi-head measurement model includes two heads. A first head correlates the measurement results of satellite signals with a confidence level of the vehicle state subjected to measurement noise. A second head correlates an estimate of the vehicle state with a confidence level of the vehicle state subjected to estimation noise. The memory also stores instructions 283 on how to determine this relationship, and stores instructions 284 to execute a probabilistic filter according to some embodiments of the invention. The memory also stores a probabilistic motion model 285 that correlates a previous confidence level of the vehicle state with a prediction of the vehicle state based on the motion model. For example, the motion model may be a vehicle dynamics model, a constant acceleration model, a constant position model, a coordinated cornering model, a Singer model, a kinematic vehicle model, a dynamic vehicle model, or a combination of different models.

[0145] In some implementations, the measurement results of the satellite signal are correlated with the confidence level of the state, and the state estimate is correlated with the confidence level of the state, by inserting the quantities into the first head and the second head of the measurement model, respectively.

[0146] The GNSS receiver is operatively connected to 250 to a GNSS measurement module, which, after determining the aforementioned delays and interference, determines the measurement results of the satellite signal and associated measurement noise. For example, in one embodiment, the measurement noise is determined based on a nominal measurement noise stored in memory 282, which is scaled as a function of the satellite's elevation angle.

[0147] The probabilistic system 200 includes an RF measurement module 240 operatively connected 250 to an RF receiver 260, which receives information 239 transmitted via an RF channel. In some embodiments, the information includes a first moment of the vehicle state. In other embodiments, the information includes a first moment and a second moment of the vehicle. In other embodiments, the information includes higher-order moments that form a general probability distribution of the vehicle state. In other embodiments, the information includes data indicating an estimate of the vehicle state and an estimate of noise, such as the first and second moments as samples, and alternatively, higher-order moments. In other embodiments, the time of receiving the information differs from the time of determining the information. For example, in some embodiments, the active remote server includes instructions for determining the first and second moments, and the execution of these instructions may be correlated with the communication time between the remote server and the RF receiver. Therefore, in some embodiments, the information 239 includes a timestamp for determining the first and second moments.

[0148] In various implementations, the GNSS measurement module and the RF measurement module generate their estimates, which relate to the confidence level of the vehicle state at the current time instance. For example, a GNSS signal is received, and upon receiving the signal, the processor 230 determines the measurement result for the current time step. However, in some implementations, the RF measurement module receives a first moment and a second moment of the vehicle state determined for a previous time step, for example, when the instructions from the remote server require time to execute. This results in the determined measurement result and the received first and second moments not corresponding to the same moment.

[0149] To address this issue, some implementations use motion models and correlate the first and second moments of the state at the previous time step with the confidence level of the state at the current time step to determine a state estimate indicating the confidence level of the vehicle's state at the current time instance. For example, one implementation uses a time propagation model of the vehicle (e.g., a motion model stored in memory) to propagate the first and second moments of the vehicle state over time up to the current time instance. Examples of time propagation models include the Singer model, constant acceleration models, and monorail vehicle models. Because of the propagation of uncertainty, using both moments in time propagation is more advantageous than propagating only the first moment; therefore, the degree to which the first moment should be trusted when updating the confidence level of the state is taken into account. In other implementations, time propagation is accomplished through a combination of the vehicle's motion model and vehicle measurements.

[0150] Some implementations are based on the understanding that, regardless of the time instance of the first and second moments of the received vehicle state, it is necessary to ensure that the estimated noise is constructed in such a way that, when the parameters of the probability distribution of the state confidence are updated with the state estimate subjected to measurement noise, the probability distribution of the state confidence converges to the probability distribution of the received state's second moment. Some implementations are based on the understanding that convergence performance can be ensured by balancing the covariance of the received second moment with the covariance of the probability distribution of the state's current confidence. In other words, the estimation of the state and the associated measurement noise is determined such that the balance remains convergent.

[0151] In some implementations, processor 230 executes a probabilistic filter 232, which is used to update the parameters of the probability distribution and determine an estimate 233 of the vehicle state.

[0152] Some implementations are based on the understanding that directly setting the parameters of the probability distribution to match the parameters of the received first and second moments may be appealing. However, enhancing convergence performance is advantageous because it ensures that, from the vehicle's perspective, the same probabilistic filter can be used to update the parameters of the probability distribution. Furthermore, since the vehicle can process additional measurements, such as those from additional onboard sensors, directly using the received first and second moments can lead to discontinuities in the updated state estimate of the vehicle. When using the estimate to control the vehicle, this discontinuity can be detrimental due to the loss of control continuity.

[0153] Figure 2C A flowchart illustrating an example method for determining an estimate of 240a according to some embodiments is shown. First, based on received information indicating an estimate of the vehicle state and the estimated noise, the method determines the distribution of the estimated vehicle state 210c. For example, the method determines the distribution by interpolating a first moment and a second moment into a Gaussian probability density function. Next, the method determines the distribution of the confidence level of the vehicle state 220c. For example, the method interpolates parameters of the probability distribution of the vehicle state, where it is assumed that the probability distribution follows a predetermined shape. For example, for a Gaussian distribution, the parameters of the probability distribution are the first moment and the second moment. For example, assuming a Student-t distribution, the parameters are the first moment, the scale, and the degrees of freedom. Using the distribution 215c of the estimated vehicle state and the distribution 225c of the confidence level of the state, the method determines the difference between distributions 215c and 225c. Finally, using the determined difference 235c, the method determines the covariance of the estimated noise 240c.

[0154] The 230c difference can be determined in several ways. For example, for a Gaussian distribution, the difference can be defined as the amount by which the second moment is scaled and the first moment is shifted to achieve the same distribution. In some implementations, the difference is determined by the well-known Kullback-Leibler divergence.

[0155] Some implementations are based on the understanding that the first two moments typically describe the distribution well, even if the underlying distribution is not Gaussian. In some implementations, this identification is used to determine the covariance of the estimated noise as a weighted combination of the second moment received by the RF measurement module and the second moment of the probability distribution of the vehicle state. In other implementations, the vehicle state estimate is determined as a weighted combination of the confidence level of the vehicle state and the first moment of the received vehicle state estimate, wherein the weighting includes the covariance of the estimated noise.

[0156] Figure 2D An example of a method according to some implementations is shown, which is used to determine a first moment and a second moment 240c to ensure that the confidence level of the vehicle state converges to the estimate of the vehicle state for a Gaussian hypothetical distribution. Using the distribution of the estimate 215c and the distribution of the confidence level 225c, the difference between the distributions is defined by a Kalman gain, which is the gain matrix required to shift the Gaussian hypothetical distribution from one distribution to another. The method starts with the distribution of the confidence level 225c and determines the Kalman gain 210d as a function of the covariance of the estimated noise, which shifts the distribution of the confidence level to the distribution determined by the received first and second moments. For example, the Kalman gain K is based on… This shifts the covariance. Based on the Kalman gain of 215d, the covariance of the estimated noise in 220d is determined by solving a matrix that includes the Kalman gain. For example, one implementation solves for the covariance of the estimated noise using the following system of equations. : This led to That is, estimating the covariance of the noise. The vehicle state estimate is determined by adjusting the confidence level of the vehicle state in response to the Kalman gain and the difference between the received first moment and the confidence level of the vehicle state. For example, the estimate is determined as... .

[0157] Some implementations use a first head and a second head of the measurement model to update the parameters of the probability distribution. For example, one implementation uses a KF with a first head of the measurement model to update the parameters of the probability distribution to adjust the parameters. Another implementation confirms that the motion model or measurement model is nonlinear for certain states of the vehicle. Therefore, some implementations employ a nonlinear KF (such as a linear regression KF) to update the parameters of the probability distribution. A linear regression KF is a filter that determines the first and second moments by performing moment integrals, where the moment integrals include a set of integration points. Evaluation and .

[0158] Some implementations recognize that one purpose of the second head of the measurement model is to allow the parameters of the probability distribution to be updated using the same principles as when using the first head of the measurement model. Some implementations recognize that in order to update the parameters of the probability distribution using an estimate of the vehicle state determined using the received first and second moments, the second head of the measurement module must ensure certain properties to ensure that the convergence of the confidence levels of the vehicle states converges to the estimate of the vehicle states. For example, in one implementation, the second head of the measurement model includes the estimated vehicle states and the confidence levels of the vehicle states determined during the previous iteration, as a difference weighted by the covariance of the estimated noise. For example, in one implementation, the second head of the measurement model is... ,in, It is an estimate of the vehicle's condition. It is the confidence level of the vehicle status, and It estimates the noise, where k is the time step in the recursion.

[0159] Some implementations are based on the understanding that, in order to update the parameters of a probability distribution using a probabilistic filter, the probabilistic filter should be able to update the parameters by taking into account the fact that part of the vehicle state is real-valued and part is integer-valued. This is because some parts (e.g., the vehicle's position) are included in the state and are real-valued, while ambiguity is included in the state and is integer-valued.

[0160] Therefore, some implementations employ probabilistic filters, which, based on... Solve a recursive mixed-integer weighted least squares problem to update the parameters of the probability distribution.

[0161] Some implementations are based on the understanding that particle filters are filters that can solve mixed-integer estimation problems without resorting to optimization methods. Other implementations understand that, in order to avoid resorting to optimization methods, particle filters require a large number of particles, which may be computationally prohibitive. Therefore, some implementations solve the recursive mixed-integer weighted least squares problem by employing a mixed-integer extended KF. Other implementations solve the mixed-integer linear regression KF.

[0162] Figure 3AA flowchart illustrating an example implementation of vehicle state estimation is shown, wherein the method receives measurement results 310a and a state estimate, including a multi-head measurement model. Each carrier signal includes carrier phase ambiguity as an unknown integer wavelength of a carrier signal traveling between satellites 101, 102, 103, or 104 and receiver 130. Next, the method retrieves from memory 322a a motion model that correlates the vehicle's previous state with its current state, and a multi-head measurement model that uses the carrier phase ambiguity of the carrier signals to correlate the measurement results of the carrier and code signals with the current confidence level of the vehicle state and correlates the estimated vehicle state with the confidence level of the vehicle state. Both models (i.e., the motion model and the measurement model) are probabilistic. For example, the motion model is a probabilistic model subjected to processing noise, while the measurement model is a probabilistic model subjected to measurement noise.

[0163] Then, based on one or a combination of the motion model and the measurement model 322a, the method determines a set of possible combinations 325a of integer values ​​of carrier phase ambiguity consistent with the measurement results of the carrier and code signals, within a range defined by one or a combination of the processing noise and the measurement noise. This step is based on the understanding that it is beneficial to identify and test different possible combinations of carrier phase ambiguities used for state estimation, rather than attempting to determine the carrier phase ambiguity to perform state estimation. In this way, the optimal carrier phase ambiguity can be selected using a probabilistic model that better reflects the nuances of state estimation.

[0164] To this end, the method performs 320a by jointly using a motion model and a measurement model 322b to determine a set of state estimators for the vehicle state. Each state estimator includes a corresponding combination of integer values ​​of carrier phase ambiguity to determine a joint probability distribution 335a of the vehicle state relative to the motion model and the measurement model 322a. In this way, the combination of integer values ​​of carrier phase ambiguity can be probabilistically evaluated because the measurement models of at least some different state estimators include different combinations of integer values ​​of carrier phase ambiguity selected from a set of possible combinations 325a. Next, the method determines 340a the vehicle state using the state estimator with the highest joint probability of the vehicle state, based on the measurements of the carrier and code signals and the estimation of the vehicle state using a multi-head measurement model.

[0165] Therefore, some implementations are based on the understanding that the estimation of the range of possible integer values ​​of carrier phase ambiguity and the selection of integer values ​​of carrier phase ambiguity from that range can be probabilistically accomplished using the consistency of the probability density function (PDF) of the noise of the first head of the motion and measurement model relative to the noise of the second head of the measurement model.

[0166] Figure 3B A flowchart is shown of a method for determining possible combinations of integer values ​​of carrier phase ambiguity 320a according to some embodiments. This method can be implemented using a processor. For at least one state of the vehicle consistent with the processing noise of the motion model, the method samples floating-point values ​​311b of the carrier phase ambiguity for at least one satellite on a PDF of measurement noise centered on a noise-free fit of the carrier phase ambiguity, converting the measurements of the carrier and code signals into estimates of the measurement model and state.

[0167] In some implementations, the PDF for measuring noise is predetermined based on, for example, the characteristics of the GNSS receiver. As used herein, noise-free fitting is the fitting of the carrier phase ambiguity, the measured carrier and code signals, and the position, assuming that the measurements of the acquired carrier and code signals are correct. That is, it is assumed that the measurements of the carrier and code signals are correct, and the residual error is due to the carrier phase ambiguity. In this way, noise-free fitting places the PDF in a position where sampling the PDF downplays the probabilistic noise of the motion and measurement model and emphasizes the effect of the carrier phase ambiguity on the position estimation. In other words, after inserting the measurements of the carrier phase ambiguity, carrier and code signals, and position into the measurement model, the error in the fitting of the measurement model is due to the error of the carrier phase ambiguity.

[0168] The method then forms a union of 320b sampled floating-point values ​​to produce a union 321b containing all sampled floating-point values; the union of floating-point values ​​is discretized 310b on an integer basis to produce a set of possible integer values ​​331b for the carrier phase ambiguity of GNSS satellites. Finally, the method uses the possible integer values ​​331b for the carrier phase ambiguity of all satellites to produce a set of possible combinations 325a of integer values ​​340b.

[0169] Figure 3C A flowchart illustrating an exemplary implementation of a method for sampling floating-point values ​​310b according to one embodiment is shown. However, sampling floating-point values ​​310b can be implemented in various ways through different embodiments. Figure 3CThe method samples floating-point values ​​consistent with the motion model and its processing noise 309c. The method determines consistency 311c with the measurement results of the code and carrier phase signals by interpolating the sampled carrier phase ambiguity, the estimated state, and the measurements of the carrier and code signals into the measurement model. Based on the consistency with the measurement results, the method corrects 312c each sampled floating-point value as a function of processing noise and measurement noise; updates the probability of each ambiguity 313c based on the consistency with the corrected measurement results; and prunes 314c the corrected sampled floating-point values ​​of the carrier phase ambiguity to maintain the floating-point values ​​of the carrier phase ambiguity, where the probability fitted to the measurement model is above a threshold.

[0170] Figure 3D Schematic diagrams illustrating some principles employed in certain implementations are shown. Specifically, some implementations are based on the understanding that a finite number of possible integer values ​​for carrier phase ambiguity allows for the determination of different combinations of these possible integer values ​​340d to track vehicle state. Such an implementation allows replacing 380d with an evaluation 310d of carrier phase ambiguity used to estimate vehicle state 330d 320d, instead of 380d, with an evaluation 360d of different vehicle states 350d determined using different combinations 340d of carrier phase ambiguity. This replacement is advantageous because the probabilistic nature of the receiver's motion is more suitable for the test position than the derivative of the test position (e.g., carrier phase ambiguity). In this way, the optimal state 370d selected using the probabilistic nature of vehicle motion automatically indicates the corresponding combination of carrier phase ambiguities used to determine this state 370d.

[0171] Some implementations are based on the understanding that particle filters may be computationally forbidden, and that Kalman filters with embedded optimization solvers may be advantageous for computational reasons.

[0172] Figure 4A flowchart of a method 260a, according to one embodiment, for performing a probabilistic filter to update the parameters of a probability distribution of a vehicle state when the parameters are the first and second moments of the distribution. The method receives 410a measurements and estimates of the state, including a multi-head measurement model. Each carrier signal includes a carrier phase ambiguity as an unknown integer wavelength of a carrier signal traveling between satellites 101, 102, 103, or 104 and vehicle 130. Next, the method retrieves 422a from memory a motion model that associates a previous vehicle state with the current vehicle state, and a multi-head measurement model that uses the carrier phase ambiguity of the carrier signals to associate the measurements of the carrier and code signals with the current confidence level of the vehicle state, and to associate the estimated vehicle state with the confidence level of the vehicle state. Both models (i.e., the motion model and the measurement model) are probabilistic. For example, the motion model is a probabilistic model subjected to processing noise, while the measurement model is a probabilistic model subjected to measurement noise. Using the parameters of the probability distribution determined during the previous iteration, the multi-head measurement model, measurement result 255a, and motion model, the method performs KF 420a to update the first and second moments of the probability distribution, where the updated first moment of the ambiguity contained in the state is real-valued. Next, the method solves a weighted least squares optimization problem 430a that fixes the ambiguity to have integer values, where the cost function of the optimization problem is the squared Euclidean norm of the deviation of the first moment of the state from the real-valued estimate 425a determined by the Kalman filter 420a. Using the obtained integer ambiguity 435a, the method then performs a Kalman filter 440a, which is initialized with the real-valued and integer-valued parts of the first moment, to obtain the updated parameters of the probability distribution 445a.

[0173] In some implementations, the KF is an extended KF, where the nonlinear portion of the motion and measurement model is linearized around the current confidence level of the state. In other implementations, the KF is a linear regression KF, such as an unscented KF, a volumetric KF, or a smart sampling KF. Linear regression KF avoids the linearization around the current state estimate as in the extended KF and is generally more accurate but computationally more complex. Instead of linearization, linear regression KF solves for the moment integrals involved through a set of weighted integration points and can also be used to determine other parameters of the probability distribution, such as higher-order moments. These higher-order moments are useful when the first and second moments cannot adequately represent the underlying distribution.

[0174] The first and second moments received from the RF receiver can be information received from multiple sources. For example, in some embodiments, a passive roadside unit (RSU) equipped with a camera uses high-accuracy computer vision algorithms to determine portions of the vehicle's state, such as position and speed. In other embodiments, the RSU is equipped with distance sensors, such as radar or lidar, which directly measure the vehicle's position and speed relative to the RSU. When partial states can be directly resolved with high accuracy using direct sensing, the RSU can determine the remaining portions of the vehicle's state and transmit them to the vehicle.

[0175] Figure 5A Communication scenarios according to some implementations are illustrated. Some implementations are based on the understanding that the active remote server resides in the cloud as an edge computing device, and that in order to establish communication between the vehicle and the cloud, information transmission between the cloud and the vehicle needs to pass through the core infrastructure network and RSU, such as... Figure 5A As shown, communication between vehicle V1 510a and cloud 520a requires the use of RSU 530a and core network 540a.

[0176] Other implementations recognize that computing in the cloud can lead to significant communication latency. In such cases, it is advantageous to have edge computing devices directly within the RSU.

[0177] Some implementations are based on the understanding that using a remote server that communicates with the vehicle requires using different clocks for different devices, and that synchronization between these clocks is necessary.

[0178] The vehicle's clock can be synchronized with the GNSS clock that communicates with the vehicle's GNSS receiver. However, some implementations are based on the understanding that, since different vehicles may have different clock offset errors, individual synchronization of the vehicle's clock with the GNSS clock cannot guarantee synchronization of the vehicle within the vehicle communication network. Furthermore, GNSS synchronization can be negatively affected by multipath propagation of satellite signals in urban environments. Additionally, some vehicles in the vehicle communication network may not have GNSS receivers.

[0179] Different standards provide protocols for synchronizing clocks. For example, IEEE has been developing synchronization standards based on the Precision Time Protocol (PTP). However, these protocols require information exchange between synchronizing devices, which may be impractical in the context of general vehicle communication networks.

[0180] Some implementations are based on the understanding that vehicular communication networks have specific requirements for participating vehicles. Unlike nodes in other networks, vehicles in vehicular communication networks periodically announce their presence. For example, in IEEE Dedicated Short Range Communication (DSRC) for Wireless Access in Vehicular Environments (WAVE), vehicles need to send a heartbeat message every 100ms to announce their presence to neighboring vehicles. The attributes of the heartbeat message include one or a combination of temporary ID, time, latitude, longitude, altitude, position accuracy, speed and transmission, heading, acceleration, steering wheel angle, braking system status, and vehicle dimensions.

[0181] Some implementations are based on the understanding that the heartbeat message includes the necessary synchronization data and can therefore be used for synchronization. This approach reduces network traffic and mitigates interference. By automatically sending synchronization data, vehicles can achieve silent synchronization without the message exchange performed in traditional synchronization methods.

[0182] For example, some implementations are based on the idea that, assuming the vehicle's clock offset and location are unknown, the vehicle's clock can be synchronized with the clocks of other vehicles based on trilateration of information received in multiple heartbeat messages. This is because vehicle trilateration can be performed in two ways. One method uses the distance light travels between the sending and receiving times of the heartbeat message, which is a function of the unknown clock offset. The other method uses the distance between the locations of multiple vehicles, the time the heartbeat message was sent, and the vehicle's unknown location at a specific time. By comparing these two distances, the vehicle's unknown location and unknown clock offset at a specific time can be determined simultaneously.

[0183] Figure 5BThis illustrates the exchange of synchronization messages in synchronization based on the IEEE Precision Time Protocol (PTP). Such synchronization requires at least two message exchanges between the master and slave devices. The master device first sends a synchronization message to the slave device at time T1. The slave device receives the Sync message at time T2. Even though the Sync message contains a timestamp T1, for more precise synchronization, the master device can also send a Follow_Up message containing the precise timestamp T1 to the slave device. After processing the Sync and Follow_Up messages received from the master device, the slave device sends a Delay_Req message to the master device at time T3. Once the master device receives the Delay_Req message at time T4, it sends a Delay_Resp message containing the timestamp T4 back to the slave device. Using the four timestamps T1, T2, T3, and T4, the slave device calculates the message propagation time and clock offset to synchronize its clock with the master device's clock. For example, for a symmetric link, the propagation time... and clock offset It can be calculated as

[0184] and

[0185] In some implementations, parameters of the probability distribution of vehicle states are sent to a remote server along with GNSS measurement results.

[0186] Figure 5C A probabilistic system 200 is illustrated, according to some embodiments, for tracking vehicle status using asynchronous cooperation of information received from satellite signals 209 transmitted from a navigation satellite system (GNSS) and information transmitted via a radio frequency (RF) channel 239. The system includes a transmitter 510c that transmits its probability distribution parameters and satellite measurement signals 511c to a remote server.

[0187] In other embodiments, multiple vehicles are transmitting parameters of their probability distributions and GNSS measurement signals to a remote server. The parameters include a first moment and a second moment of the probability distribution of the vehicle states. This information transmission can be beneficial when the remote server (e.g., located in an RSU or in the cloud) does not have additional sensing capabilities. In this case, the remote server can use information received from multiple vehicles to combine this information, thereby improving the estimation of the state of each vehicle, and transmit the first and second moments to the vehicles. In other words, the probabilistic system of the vehicles transmits parameters of its probability distributions along with satellite measurements, and in response to transmitting the first and second moments of the received vehicle states. In some embodiments, the remote server forms an enhanced state based on the states of multiple vehicles.

[0188] In some implementations, the enhanced state is the union of the states of multiple vehicles. For example, the enhanced state includes vehicle position and velocity, carrier phase ambiguity, and residual states, such as ionospheric residual delay. For vehicles that are sufficiently close to each other (e.g., vehicles less than 3-10 km apart), the ionospheric delay can be considered the same for different vehicles. For example, refer to... Figure 5D Vehicles within region 510d exhibit very similar ionospheric delays. Therefore, in some implementations, the dimension of the enhanced state is less than the sum of the states of all vehicles, since such ionospheric delays are identical for multiple vehicles.

[0189] In most GNSS applications, residuals are suppressed through single-difference or double-difference operations, with the focus on recovering integer ambiguities that form part of the GNSS measurement results. However, some implementations are based on the understanding that residuals still affect GNSS measurement results due to the physical nature of GNSS signal propagation. These biases, however, cannot be determined solely from a single GNSS receiver, but rather through the enhancement of measurements from multiple satellites measured from multiple receivers. This is because integer ambiguities are unique for each satellite-vehicle combination, but residuals are not. For example, ionospheric delay is included as a residual in the state, and the same applies to vehicles measured near each other. Therefore, to recover these residuals, measurement noise should be estimated to determine the correlation between measurements of multiple vehicles from satellite signals—a task quite different from integer ambiguity estimation.

[0190] Some implementations are based on the understanding that estimating residuals is difficult to solve at the level of a single vehicle due to numerous unknowns. This is because, while each vehicle receives multiple GNSS signals from multiple satellites, and the residuals of different GNSS signals are similar, the ambiguity is not. Therefore, even for a single receiver, using more measurements does not significantly help in estimating residuals.

[0191] On the other hand, when multiple measurements from multiple vehicles are used, there is cross-correlation between the GNSS measurements from different vehicles because these measurements come from the same satellite.

[0192] Figure 6A A flowchart of a recursive method 600 for cooperative state tracking of multiple vehicles, according to some embodiments, is shown. The method is executed by a processor of a remote server (e.g., an RSU or an edge device in the cloud). The remote server receives 610a multiple transmissions 605a from the multiple vehicles. Each transmission 605a from the vehicles includes measurements of satellite signals received at the vehicle and parameters of the probability distribution of the vehicle's state.

[0193] A remote server merges the states of multiple vehicles defined by parameters of corresponding probability distributions to form an enhanced state 625a of multiple vehicles 620a. In one implementation, method 600 simply concatenates all states together in a vector, matrix, or tensor. However, in some other implementations, some residuals of adjacent vehicles are identical, and method 600 removes the copy that forms the enhanced state 625a into a union of the states of multiple vehicles 620a, such that the dimension of the enhanced state is less than the sum of the dimensions of the states of each of the multiple vehicles.

[0194] Furthermore, method 600 fuses 630a the measurement results of satellite signals from multiple vehicles into an enhanced measurement result subjected to enhanced measurement noise 635a. Some implementations are based on the dependence of measurement noise on different vehicles. Therefore, the enhanced measurement noise 623a is defined by an off-diagonal covariance matrix with non-zero off-diagonal elements, each of which is related to the error in the measurement results of a pair of corresponding vehicles.

[0195] Method 600 performs a probabilistic filter update of the enhanced state 645a based on the enhanced measurement results subjected to enhanced measurement noise. Due to one or a combination of redundancy of state variables and cross-correlation of measurement noise, the accuracy of the updated enhanced state 645a for each individual vehicle can be greater than the accuracy of individually determined vehicle states.

[0196] To avoid control discontinuities, method 600 fuses the parameters of the probability distributions of the states of at least some of the vehicles with the corresponding portions of the updated enhanced state 645a, 650a, to output fused parameters 655a of the probability distributions of the states of at least some of the vehicles. The fused parameters 655a can be sent to the vehicles for vehicle control and for other purposes utilizing state estimation.

[0197] The implementation of probabilistic filters 640a can be performed in a variety of ways using different measurement models (such as multi-head or single-head measurement models) and various motion models (such as constant velocity motion models, constant acceleration motion models, Singer models, moving vehicle models, and dynamic vehicle models). Examples of probabilistic filters include particle filters or mixed-integer Kalman filters.

[0198] Some implementations are based on the understanding that the states of multiple vehicles can be jointly tracked using probabilistic estimation principles similar to those used for probabilistic tracking of a single vehicle. At first glance, such joint tracking seems pointless, as it increases the dimensionality of the tracking problem without immediately providing any significant benefits. This is because the integer ambiguities of different vehicles (which are the focus of most GNSS applications) are independent of each other in such collective tracking, and therefore collective tracking will not help resolve ambiguities.

[0199] However, some implementations are based on the understanding that during the collective state estimation of multiple vehicles, the measurement noise applied to the collective state of the multiple vehicles (i.e., referred to as the enhanced state in this disclosure) does indeed have a useful cross-correlation between the noise of the measurement results of the different vehicle states, which can be explored to reduce the uncertainty of the enhanced state estimation. For example, when a satellite measures two vehicles, there is a direct cross-correlation between the two measurements because they originate from the same source and therefore have similar noise characteristics.

[0200] Some implementations are based on the understanding that augmented state estimation can correct or at least improve state estimation performed by a single vehicle. However, it is necessary to soft-merge the results of augmented state tracking and individual state tracking to avoid discontinuities in vehicle control. This is because a single vehicle can be updated at a much faster rate and has additional onboard sensors, such as inertial measurement units and wheel encoders, which can provide different information. Therefore, simply using the results from the augmented state directly can lead to a loss of continuity in the estimation, which could be detrimental to vehicle safety. To this end, some implementations fuse the parameters of the probability distribution of the individual vehicle's state with the corresponding portion of the updated augmented state, rather than updating the state of the individual vehicle itself. In this way, the resulting estimate is a trade-off between additional information (such as associated measurement noise and the similarity of residuals used by the server) and the smoothness provided by always using individual vehicle estimates.

[0201] Alternatively or concurrently, some implementations recognize the need to reduce the computational burden of such high-dimensional probabilistic filters for tracking augmented states. For example, a vehicle's state can have a state dimension as high as 100, depending on how many residuals are estimated and how many satellites are currently measuring the vehicle. On the other hand, there may be hundreds of vehicles nearby, and in the worst case, the dimension of the augmented state could be tens of thousands. Therefore, the computations involved in remote servers become computationally demanding, especially the solution to the mixed-integer estimation problem, which requires fixing a large number of integer ambiguities, becomes daunting. Some implementations recognize that mixed-integer Kalman filters should be used to solve the mixed-integer least-squares problem for augmented states.

[0202] Some implementations are based on the understanding that designing a recursive mixed-integer estimator means solving a mixed-integer problem of ever-increasing size, since the current estimate of the state depends on the entire history of integer values. Therefore, to solve such a large-scale estimation problem, various relaxations are required, because otherwise the complexity would grow exponentially over time.

[0203] One implementation is based on the understanding that, rather than determining integer ambiguities by optimizing from real-valued ambiguities and subsequently updating the parameters of the distribution based on said integer ambiguities, it is advantageous to recursively and jointly determine the parameters of two probability distributions, where one distribution applies only to real-valued states and the other models integer ambiguities. This is because if two distributions are used, the real-valued estimation is not mixed with the mixed-integer estimation, and therefore the true distribution of the real-valued states is not contaminated.

[0204] Figure 6B A flowchart of a recursive method for updating the parameters of the probability distribution of the augmentation state of 620a at a remote server, according to some embodiments, is shown. Using measurements from multiple transmissions 605a, the method jointly updates two sets of parameters of the distribution, one set of parameters without real-value constraints and the other set of parameters constrained to make the ambiguity integer values. First, using the motion model and measurement model 622b, the method determines a one-step prediction of the parameters of the probability distribution of the unconstrained augmentation state of 620b. Next, the method updates the parameters of the probability distribution of the unconstrained augmentation state of 630b using satellite measurements.

[0205] Using real-valued parameters of the ambiguity, this method fixes the 640b ambiguity and uses constrained ambiguity and measurement model 622b. This method generates a constraint relationship by determining the relationship between satellite measurement results and augmentation states with integer ambiguities, thereby constraining the 619b augmentation state. Next, the method determines a one-step prediction of the parameters of the probability distribution of the constrained augmentation state 639b, and finally updates the parameters of the probability distribution of the constrained augmentation state 649b.

[0206] One-step predictions can be made using a motion model and a measurement model that are jointly updated using a Kalman filter. Parameter updates can be similarly accomplished using a Kalman filter. For example, some implementations use an extended Kalman filter to perform time updates, i.e., one-step predictions are...

[0207] Some implementations use a weighted least squares problem to constrain the ambiguity constraint 640b to an integer value. Wherein, the weights are the updated second moments of the unconstrained augmentation state, and where, It is the first moment of the real-valued (i.e., unconstrained) ambiguity.

[0208] Some implementations are based on the understanding that ambiguities exhibit jumping behavior due to occasional cycle slips. During a cycle slip event, an ambiguity may remain unchanged for a long period before a single ambiguity suddenly jumps to a new integer value. Some implementations detect cycle slips by modeling the ambiguity's temporal evolution (i.e., ambiguity motion model 622b) as a discrete random walk. In some implementations, for ambiguity n, the ambiguity's temporal evolution is modeled as a discrete random walk. .

[0209] in, It is the variance reflecting the field width of integer jump processing. It is a regularization term, and Determine if a cycle slip has occurred.

[0210] In some implementations, the difference between the satellite measurement results and the predicted satellite measurement results is formed to determine whether a cycle slip has occurred. For example, one implementation forms the difference as... Furthermore, it was recognized that the most likely event causing a sudden change in the difference between two consecutive timestamps is due to a cycle slip. Therefore, the difference equals This means any dimension of the difference in ambiguity. Sufficiently far from the origin, determined by a threshold d, due to cycle slips. In one implementation, the occurrence rate... identified as .

[0211] Figure 6C A flowchart of a method 630a for fusing multiple states of a vehicle with an enhanced state, according to some embodiments, is shown. Since the vehicle's states and the enhanced state typically do not have the same dimensions, it is necessary to transform the states into a joint state space that equally represents the different states. In one implementation, method 630a transforms 610c the parameters of the vehicle states to the same state space as the parameters of the enhanced state.

[0212] In one implementation, for an enhanced state distribution having a first moment and a second moment and vehicle status , conversion by Definition, which means two sets of first and second moments defined in the augmented state space. , , , In one implementation, the state spaces are equal, and such a transformation is not required. Using the transformed parameters, the method fuses 620c to form an updated set of parameters for the probability distribution.

[0213] Fusion can be performed in several ways. In one implementation, parameter fusion 620c is accomplished as a weighted combination between the enhanced state and the state of each vehicle. For example, one implementation fuses the first and second moments into a convex combination of moments. The weights are determined by minimizing the determinant of the fusion covariance. In other implementations, weights are determined by minimizing the trajectory of the fusion covariance. In another implementation, the weights are determined by setting weights based on a ratio. In another different implementation, fusion is performed based on a weighted combination. .

[0214] Figure 6D An example of overlapping state spaces according to some implementations is shown. Figure 6D In the left-hand portion, the state spaces of the three different vehicles 610d, 620d, and 630d, and the enhanced state 640d, are similar, meaning no transition is required between them. Figure 6D In the middle part of the diagram, the state spaces are completely disjoint, while in the right diagram, there is overlap between the states.

[0215] The above fusion rules are conservative because they do not take into account the cross-variance between the augmented state and the vehicle state.

[0216] Figure 6E A flowchart of a method 620c for fusing enhanced state with vehicle state according to some embodiments is shown. If necessary, such as… Figure 6D In the third case, the method uses the combined state space of the transformed 610e parameters and the measurement model 615e based on satellite measurements to recursively update the cross-covariance between vehicle states and between vehicle states and augmentation states. Finally, the method fuses the first and second moments of 630e based on the determined cross-covariance.

[0217] The recursive determination of the cross-covariance 620e can be accomplished in several ways. For example, one implementation recognizes that if a Kalman filter, such as an extended Kalman filter, a linear regression Kalman filter, a mixed-integer extended Kalman filter, or a mixed-integer linear regression Kalman filter, is used, the cross-covariance can be recursively determined at each estimated time step k. This is because the probability distribution in such a filter is represented by the first two moments of a Gaussian distribution.

[0218] Figure 6FA flowchart of a method 620c for determining cross-covariance according to one embodiment is shown. Using a motion model and a measurement model 615f of the enhanced state and vehicle state, the method determines the Kalman gain of the Kalman filter involved in 610f. For example, one embodiment determines the Kalman gain of the vehicle and estimates the Kalman gain of the Kalman filter for the enhanced state, wherein the Kalman gain is determined according to a recognized technique.

[0219] One implementation is based on the understanding that if a Kalman filter is used, the probability distribution is of Gaussian form. ,in, It is the first moment, and It is the second moment. If a Kalman filter is used, the measurement result is used. Update the first two moments using Kalman gain Complete. For a set of N vehicles, one implementation stacks the vehicle transition states into a state vector. In one implementation, the cross-covariance between vehicle states is determined as... .

[0220] in, It is the cross-covariance, which is recursively determined as a function of the Kalman gain, and where and Examples are given to illustrate the motion model and measurement model matrices, respectively.

[0221] Some implementations employ stacked motion models, where the motion models can be stacked vehicle dynamics models, stacked constant acceleration models, stacked constant position models, stacked coordinated turning models, stacked Singer models, or combinations of different stacked models. Stacking refers to the mutual stacking of multiple models. For example, for a vehicle motion model...

[0222] in, It refers to the number of vehicles. It refers to motion state and ambiguity, and It is a residual state. This refers to the enhanced state of the motion model, which handles noise and stacking. Similarly, the GNSS measurement model can be defined based on the enhanced state. For example, for a set of measurement results... , A complete measurement model includes measurement results. .

[0223] Another implementation determines the cross-covariance between the vehicle state and the enhanced state of the remote server as a function of the joint Kalman gain of the vehicle state. And determine the Kalman gain of the Kalman filter on the remote server as a combination. .

[0224] Alternatively or additionally, the Kalman gain is a portion of the information 215a received from the vehicle.

[0225] Other implementations are based on the understanding that recursively calculating the cross-covariance can be problematic when using a particle filter because, in this approach, the distribution is represented not only by parameters as the first and second moments, but also by a mixture of the first and second moments as a Gaussian mixture. However, other implementations recognize that, for a particle filter, the Kalman gain of the most probable particle can be used alternatively.

[0226] Using the determined cross-covariance of 620e, method 620c fuses the parameters of the probability distribution of the vehicle state with the parameters of the probability allocation of the enhanced state in the remote server 630e. In one embodiment, the fusion of the first moment and the second moment is accomplished as a weighted combination of the first moment and the second moment involved, respectively. For example, the fused first moment is determined based on the weighted difference between the first moments, where the weights are the cross-covariance. Weighted combination of individual covariances Another implementation determines the fused covariance based on the weighted difference between the second moments, where the weights are the cross-covariances. Weighted combination of individual covariances Alternatively or otherwise, the first and second moments of fusion are determined as follows: .

[0227] One implementation recognizes that the fusion of parameters involves several inverses of potentially very high-dimensional matrices. One implementation uses Schur complement to reduce the number of inversions required to produce the fused covariance. The size of the matrix

[0228] Other implementations are based on the understanding that when using measurement results from multiple vehicles, additional information can sometimes be inferred from the measurement results.

[0229] Figure 6G An example of cross-covariance in measurement noise of a full measurement model according to some implementations is shown, which includes measurement results from multiple transmissions from multiple vehicles. Figure 6GThe matrix showing the elements representing the covariance of measurement noise is illustrated. When stacked measurements result in an enhanced measurement model for the enhanced state, but the satellites are not shared between vehicles, the diagonal 610g will be the only part with a non-zero component. However, when vehicles share the same satellite used to create differential satellite measurements (i.e., single-difference or double-difference), the deviations from the diagonals 620g and 630g can have a non-zero component 640g, if correctly modeled. Probabilistic filters utilize this non-zero component to increase knowledge of the enhanced state estimate.

[0230] Some implementations understand that when the measurement noise matrix R has been formed, it can be automatically used in the Kalman Flow (KF) regardless of the presence of off-diagonal non-zero elements. However, when there are non-zero off-diagonal elements (i.e., non-zero cross-covariance), the KF can utilize this to better determine the Kalman gain, which is based on the measurement error to determine the update of the first and second moments.

[0231] In one implementation, using the principle that creates a covariance between the measurement results, the measurement noise matrix is ​​formed by single and / or double differences of the code and carrier phase measurements. For example, in some implementations, the non-zero off-diagonal element 640 corresponds to a vehicle pair and is determined by a combination of the covariances of the corresponding vehicle pairs.

[0232] Figure 6H The method for determining the measurement noise matrix to generate it is shown. Figure 6G The flowchart describes the method for sorting satellite measurement results (610g) to form a structure of sorted satellite measurement results (615h). For example, the measurement results are sorted according to the satellite's elevation angle, where the satellite with the highest elevation angle is ranked first, the satellite with the second highest elevation angle is ranked second, and so on, until the satellite with the lowest elevation angle is ranked last.

[0233] Next, this method calculates the difference between the 620-hour satellite measurement results. This can be achieved in various ways, but it is important that it is done in a way that makes the correlation visible. One implementation defines a binary correlation matrix for vehicle i. The binary association matrix is ​​a matrix from which elements are extracted from a vector. For example, let... It is the residual vector consisting of all unique deviations, and let It is a set of satellites targeting vehicle i The corresponding deviations, among which, It is all possible visible satellites and The satellite was observed through a proxy AI. The set, This is the set of satellites visible to vehicle i. Another implementation defines a difference operator. Using a binary correlation matrix and difference operators, one implementation forms the difference between satellite measurements as... ,in, It is based on all codes and carrier phase measurement results The set determines the code of vehicle i and the carrier measurement results.

[0234] Other implementations form the difference as a double difference for vehicle i. In one implementation, the defined binary incidence matrix and difference operator are used to rewrite the difference as... .

[0235] Another implementation models the nominal code and carrier noise as Gaussian random variables with covariance, when hour, and In other words, double difference can be represented using the binary incidence matrix, the difference operator, and nominal noise. Using a difference of 625h, this method determines that 630h has the following characteristics: Figure 6G The measurement noise matrix with non-zero off-diagonal elements is shown.

[0236] In one implementation, a measurement noise matrix with non-zero off-diagonal elements of 630h is determined by selecting satellites with the same base station during the differencing process. In another implementation, a covariance is formed between vehicle i and vehicle j, and this results in… This is used for carrier phase measurement results and similarly for code measurement results. In other words, by determining the difference compared to satellites at the same base station, the additional information contained in the satellite measurements can be used to update the first and second moments using KF enhancement, since the satellites are shared among vehicles, i.e., one satellite measures several vehicles.

[0237] Figure 7 A server 700 is shown that, according to some embodiments, uses asynchronous cooperation to track the status of multiple vehicles using information received from vehicles (e.g., information transmitted via RF channel 709). The server includes an RF receiver 760 for receiving multiple transmissions of first and second moments from the multiple vehicles, satellite measurement results, and corresponding measurement noise of the measurement results.

[0238] The system includes a memory 780 that stores a probabilistic measurement model 781 that associates incoming satellite measurements with vehicle states. The memory also stores a probabilistic motion model 782 that associates a vehicle's previous state with a prediction of its state based on the motion model. For example, the motion model could be a stacked vehicle dynamics model, a stacked constant acceleration model, a stacked constant position model, a stacked coordinated turning model, a stacked Singer model, or a combination of different stacked models, where stacking means multiple models are stacked on top of each other.

[0239] The memory also stores instruction 783 to execute a probabilistic filter according to some embodiments of the present invention.

[0240] The probabilistic system 700 includes an RF measurement module 740 operably connected 750 to an RF receiver 760, which receives information 709 transmitted from multiple vehicles via an RF channel. In some embodiments, this information includes first and second moments of the multiple vehicles. In other embodiments, this information includes satellite signals received from the vehicles, wherein these signals include satellite measurement results, position information, and timing information.

[0241] Some implementations use a motion model to correlate the first and second moments of the state at a previous time step with the first and second moments of the state at the current time step. For example, one implementation uses a time propagation model that leverages the motion model to propagate the first and second moments of the vehicle state over time up to the current time instance. Examples of time propagation models include the Singer model, the constant acceleration model, and the monorail vehicle model. Because of the propagation of uncertainty, using both moments in time propagation is more advantageous than propagating only the first moment; therefore, the degree to which the first moment should be trusted when updating the state is taken into account. In other implementations, time propagation is accomplished through a combination of the vehicle's motion model and vehicle measurements.

[0242] In various implementations, the probabilistic system uses a first moment and a second moment, along with satellite measurements, to update the first moment and the second moment using information received from multiple transmissions from multiple vehicles. For example, a GNSS signal is received, and upon receiving the signal, the processor 730 uses a probabilistic filter 732 to determine the updated first moment and the second moment 733.

[0243] In other embodiments, the probabilistic system 700 includes an RF transmitter 770 for transmitting 739 updated first and second moments (i.e., mean and covariance) corresponding to a vehicle. For example, the mean and covariance of vehicle 1 are transmitted to vehicle 1.

[0244] Exemplary Implementation

[0245] Figure 8 An example of state-estimated vehicle-to-vehicle (V2V) communication and planning according to one implementation is shown. As used herein, each vehicle can be any type of mobile transportation system, including buses, mobile robots, or rovers. For example, the vehicles can be autonomous or semi-autonomous.

[0246] In this example, multiple vehicles 800, 810, and 820 are traveling on a given highway 801. Each vehicle can perform various actions. For example, vehicles can remain on the same paths 850, 890, and 880, or change paths (or lanes) 860 and 870. Each vehicle has its own sensing capabilities, such as lidar, cameras, etc. Each vehicle can have the ability to send and receive information 830 and 840 with its neighboring vehicles and / or indirectly exchange information with other vehicles via a remote server. For example, vehicles 800 and 880 can exchange information through vehicle 810. Using this type of communication network, information can be transmitted over a large portion of the toll-free highway or highway 801.

[0247] Some implementations are configured to address situations where, for example, vehicle 820 wants to change its path and selects option 870 in its path planning. However, simultaneously, vehicle 810 also selects to change its lane and wants to follow option 860. In this situation, a collision between the two vehicles is possible, or at best, vehicle 810 will have to perform emergency braking to avoid colliding with vehicle 820. This is where the present invention can help. To this end, some implementations enable the vehicle not only to transmit what it senses at the current time t, but also, additionally or alternatively, to transmit what it is planning to do at a given time. do what.

[0248] exist Figure 8 In the example, vehicle 820 notifies vehicle 810 of its lane-changing plan after planning and committing to executing it. Therefore, vehicle 810 knows that... During the time interval, vehicle 820 is planning to move to the left at 870. Therefore, vehicle 810 can choose to move at 890 instead of 860, that is, stay in the same lane.

[0249] Alternatively or concurrently, the remote server may jointly control the movement of the vehicles based on state estimates determined collaboratively. For example, in some implementations, the multiple vehicles determined for the joint state estimate are vehicles that form, and may form, a vehicle platoon jointly controlled using a shared control objective.

[0250] Figure 9This is a schematic diagram of multi-vehicle platooning formation for an accident avoidance scenario according to one embodiment. For example, consider a group of vehicles 930, 970, 950, and 960 traveling on highway 901. Now consider that suddenly, an accident occurs ahead of the platoon in area 900. This accident makes area 900 unsafe for vehicle traffic. Vehicles 920 and 960 sense the problem, for example, using cameras, and transmit this information to vehicles 930 and 970. The platoon then executes a distributed optimization algorithm, such as a formation-keeping multi-agent algorithm, which selects the optimal shape of the platoon to avoid the accident area 900 and maintain uninterrupted traffic flow. In this illustrative example, the optimal shape of the platoon is to align and form a line 995 that avoids area 900.

[0251] Figure 10 A block diagram of a system 1000 for direct and indirect control of hybrid autonomous vehicles, according to some embodiments, is shown. System 1000 may be deployed as part of an RSU on a remote server to control passing hybrid autonomous vehicles, including autonomous, semi-autonomous, and / or manually driven vehicles. System 1000 may have multiple interfaces for connecting system 1000 to other machines and devices. A network interface controller (NIC) 1050 includes a receiver adapted to connect system 1000 to a network 1090 connecting system 1000 and hybrid autonomous vehicles via bus 1006 to receive traffic conditions of a group of hybrid autonomous vehicles traveling in the same direction, wherein the group of hybrid autonomous vehicles includes controlled vehicles willing to participate in platooning and at least one uncontrolled vehicle, and wherein the traffic conditions indicate the state of each vehicle and controlled vehicle in the group. For example, in one embodiment, the traffic conditions include the current headway, current speed, and current acceleration of the hybrid autonomous vehicles. In some embodiments, the hybrid autonomous vehicles include all uncontrolled vehicles within a predetermined range from the flanking controlled vehicles in the platoon.

[0252] The NIC 1050 also includes a transmitter adapted to send control commands to the controlled vehicles via network 1090. For this purpose, system 1000 includes an output interface, such as control interface 1070, configured to submit control commands 1075 to the controlled vehicles in the group of hybrid autonomous vehicles via network 1090. In this way, system 1000 can be deployed on a remote server that communicates directly or indirectly with the hybrid autonomous vehicles.

[0253] System 1000 may also include other types of input and output interfaces. For example, system 1000 may include a human-machine interface 1010. The human-machine interface 1010 can connect the controller 1000 to a keyboard 1011 and a pointing device 1012, wherein the pointing device 10120 may include a mouse, trackball, touchpad, joystick, pointer, stylus, or touch screen, etc.

[0254] System 1000 includes a processor 1020 configured to execute stored instructions and a memory 1040 storing instructions executable by the processor. Processor 1020 may be a single-core processor, a multi-core processor, a computing cluster, or any number of other configurations. Memory 1040 may include random access memory (RAM), read-only memory (ROM), flash memory, or any other suitable memory machine. Processor 1020 may be connected to one or more input and output devices via bus 1006.

[0255] Processor 1020 is operatively connected to storage device 1030, which stores instructions and processing data used by the instructions. Storage device 1030 may form part of or be operatively connected to memory 1040. For example, memory may be configured to store probabilistic filter 1031 trained to track the enhanced state of a hybrid autonomous vehicle, converting traffic states into target inter-vehicle distances for the hybrid autonomous vehicle; and to store one or more models 1033 configured to interpret vehicle motion. For example, model 1033 may include motion models, measurement models, traffic models, etc.

[0256] Processor 1020 is configured to determine control commands for controlled vehicles, which also indirectly control uncontrolled vehicles. To this end, the processor is configured to execute control generator 1032 to determine control commands based on vehicle states. In some embodiments, control generator 1032 uses a deep reinforcement learning (DRL) controller trained to generate control commands for individual vehicles and / or convoys from reinforced states.

[0257] Figure 11A A schematic diagram of a vehicle 1101, directly or indirectly controlled according to some embodiments, is shown. As used herein, vehicle 1101 can be any type of wheeled vehicle, such as a bus, public transport vehicle, or mobile vehicle. Furthermore, vehicle 1101 can be automatic or semi-automatic. For example, some embodiments control the movement of vehicle 1101. Examples of movement include lateral movement of the vehicle controlled by a steering system 1103 of vehicle 1101. In one embodiment, steering system 1103 is controlled by a controller 1102 communicating with system 1000. Alternatively or additionally, steering system 1103 may be controlled by the driver of vehicle 1101.

[0258] The vehicle may also include an engine 1106, which may be controlled by the controller 1102 or other components of the vehicle 1101. The vehicle may also include one or more sensors 1104 to sense the surrounding environment. Examples of sensors 1104 include rangefinders, radar, lidar, and cameras. The vehicle 1101 may also include one or more sensors 1105 to sense its current motion and internal state. Examples of sensors 1105 include Global Positioning System (GPS), accelerometers, inertial measurement units, gyroscopes, shaft rotation sensors, torque sensors, deflection sensors, pressure sensors, and flow sensors. The sensors provide information to the controller 1102. The vehicle may be equipped with a transceiver 1106 that enables the controller 1102 to communicate via wired or wireless communication channels.

[0259] Figure 11B A schematic diagram illustrates the interaction between a controller 1102 receiving controlled commands from system 1000 and a controller 1100 of vehicle 1101, according to some embodiments. For example, in some embodiments, the controller 1100 of vehicle 1101 is a steering 1110 and a brake / throttle controller 1120 that control the rotation and acceleration of vehicle 1100. In this case, controller 1102 outputs control inputs to controllers 1110 and 1120 to control the vehicle state. Controller 1100 may also include an advanced controller, such as a lane-keeping assist controller 1130, which further processes the control inputs of predictive controller 1102. In both cases, controller 1100 maps the output of predictive controller 1102 to control at least one actuator of the vehicle (such as the vehicle's steering wheel and / or brakes) to control the vehicle's motion. This can include position, orientation, and longitudinal / lateral velocity; control input. This may include lateral / longitudinal acceleration, steering angle, and engine / brake torque. The system's state constraints may include lane-keeping constraints and obstacle avoidance constraints. Control input constraints may include steering angle constraints and acceleration constraints. Collected data may include position, orientation, and speed profiles, acceleration, torque, and / or steering angle.

[0260] The embodiments of the present invention described above can be implemented in various ways. For example, the embodiments can be implemented using hardware, software, or a combination thereof. When implemented in software, the software code can be executed on any suitable processor or set of processors, whether provided in a single computer or distributed among multiple computers. Such a processor can be implemented as an integrated circuit having one or more processors within an integrated circuit assembly. However, the processor can be implemented using circuitry of any suitable format.

[0261] Furthermore, the various methods or processes outlined herein can be encoded as software executable on one or more processors, utilizing any of a variety of operating systems or platforms. Moreover, such software can be written using any of a variety of suitable programming languages ​​and / or programming or scripting tools, and can also be compiled into executable machine language code or intermediate code that executes on a framework or virtual machine. Typically, in various implementations, the functionality of program modules can be combined or distributed as needed.

[0262] Furthermore, embodiments of the present invention can be specifically implemented as a method, examples of which have been provided. Actions performed as part of the method can be ordered in any suitable manner. Therefore, embodiments can be configured to perform actions in an order different from that shown, which may include performing some actions simultaneously, even if these actions are shown as sequential actions in the illustrated embodiments.

[0263] Although the invention has been described by way of examples of preferred embodiments, it should be understood that various other changes and modifications can be made within the spirit and scope of the invention. Therefore, the appended claims are intended to cover all such changes and modifications within the true spirit and scope of the invention.

Claims

1. A probabilistic system that tracks vehicle status using asynchronous cooperation of information received from satellite signals transmitted from a Global Navigation Satellite System (GNSS) and information transmitted via a radio frequency (RF) channel, said probabilistic system comprising: A memory configured to store a probabilistic multi-head measurement model that associates incoming measurement results with the vehicle state, wherein the probabilistic multi-head measurement model includes a first head and a second head, the first head associating measurement results of satellite signals subjected to measurement noise with a confidence level of the vehicle state, and the second head associating an estimate of the vehicle state subjected to estimation noise with a confidence level of the vehicle state; and At least one processor, configured to process executable instructions of a module of the probabilistic system, the module comprising: A GNSS measurement module, operatively connectable to a GNSS receiver and configured to determine the measurement results of the satellite signal and the measurement noise; An RF measurement module, operatively connectable to the vehicle's RF receiver and configured to receive data indicating the vehicle's state and the estimated noise; and A probabilistic filter is configured to recursively update the parameters of the probability distribution of the vehicle state to generate a confidence level of the vehicle state based on a multi-head measurement model that receives measurement results of the satellite signal subjected to the measurement noise and estimates of the vehicle state subjected to the estimation noise, or a combination thereof.

2. The probabilistic system according to claim 1, wherein, The probabilistic filter recursively propagates the parameters of the probability distribution of the vehicle state based on the motion model of the vehicle state transition subjected to processed noise, and updates the parameters of the probability distribution upon receiving the output of one or a combination of the first head and the second head of the multi-head measurement model.

3. The probabilistic system according to claim 1, wherein, The vehicle state includes the ambiguity of the satellite signal propagation and the state deviation of the residuals in one or more atmospheric capture delays.

4. The probabilistic system according to claim 1, wherein, Upon receiving the measurement results from the satellite signal and the measurement noise, the probabilistic filter executes the first head of the multi-head measurement model to: The vehicle state is estimated based on the measurement results that conform to the satellite signal, thereby generating the measured vehicle state; The measurement noise is converted into a second moment representing the covariance of the measured vehicle state; as well as The confidence level of the vehicle state is compared with the measured vehicle state using the covariance measurement to output the comparison result.

5. The probabilistic system according to claim 4, wherein, The output of the first head of the multi-head measurement model includes the error between the measured vehicle state and the current confidence level of the vehicle state inserted into the measurement model subject to the covariance of the measurement noise.

6. The probabilistic system according to claim 4, wherein, The output includes the measured vehicle state, the current confidence level of the state of the first head of the multi-head measurement model inserted into the measurement model, and the covariance of the measurement noise.

7. The probabilistic system according to claim 4, wherein, Upon receiving an estimate of the vehicle state and the estimated noise, the probabilistic filter performs a second head operation on the multi-head measurement model to: The received estimate of the vehicle state is converted into the estimated vehicle state; The received estimated noise is converted into the estimated vehicle state covariance that conforms to the measured vehicle state covariance. as well as The current confidence level of the vehicle state is compared with the estimated vehicle state estimated using the covariance of the estimated vehicle state, and the comparison result is output.

8. The probabilistic system according to claim 7, wherein, The received estimate of the vehicle state is a first moment of the vehicle state, and the received estimated noise is a second moment of the vehicle state.

9. The probabilistic system according to claim 8, wherein, The estimated covariance of the vehicle state is determined as a weighted combination of the second moment of the estimated vehicle state and the second moment of the probability distribution of the confidence level of the vehicle state.

10. The probabilistic system according to claim 8, wherein, The estimated covariance of the vehicle state is determined based on the difference between the probability distribution defined by the first and second moments of the received estimated vehicle state and the probability distribution of the confidence of the vehicle state defined by the parameters maintained by the probabilistic filter.

11. The probabilistic system according to claim 8, wherein, The estimated vehicle state is determined as a weighted combination of the confidence level of the vehicle state and the first moment of the received vehicle state, wherein the weighting includes the covariance of the estimated noise.

12. The probabilistic system according to claim 2, wherein, The probabilistic filter propagates the parameters of the probability distribution of the vehicle state at each time instance, and updates the parameters of the probability distribution of the vehicle state for the current time instance based on the output of the first head or the output of the second head of the multi-head measurement model.

13. The probabilistic system according to claim 2, wherein, The probabilistic filter propagates the parameters of the probability distribution of the vehicle state at each time instance, and updates the parameters of the probability distribution of the vehicle state for the current time instance based on the combination of the outputs of the first head and the second head of the multi-head measurement model, which are incorporated into the enhanced output of the multi-head measurement model.

14. The probabilistic system according to claim 1, wherein, The received data indicates the estimated noise and the estimated vehicle state for previous time instances, and wherein, in order to update the parameters of the probability distribution of the vehicle state in the current time instance, the estimated noise and the estimated vehicle state for previous time instances are propagated using a time propagation model up to the current time instance.

15. The probabilistic system according to claim 14, wherein, The time propagation model includes one or a combination of the vehicle's motion model, the Singer model, the constant acceleration model, and the monorail vehicle model.

16. The probabilistic system according to claim 1, wherein, The probabilistic filter is one or a combination of a mixed-integer extended Kalman filter, a mixed-integer linear regression Kalman filter, a linear Kalman filter, an extended Kalman filter, and a particle filter.

17. The probabilistic system according to claim 1, wherein, The RF measurement module receives an estimate of the vehicle state and the estimated noise from the roadside unit (RSU).

18. The probabilistic system according to claim 1, further comprising: A transmitter configured to transmit parameters of a probability distribution of the vehicle state to a remote server, the parameters including at least a first moment and a second moment of the probability distribution and a measurement result of the satellite signal, such that the RF measurement module, in response to the transmission, receives an estimate of the vehicle state and the estimated noise.

19. The probabilistic system according to claim 18, wherein, The remote server is configured as follows: Receive multiple transmissions from multiple vehicles, the multiple transmissions including transmissions from the vehicles, each transmission from a respective vehicle including measurement results of satellite signals and parameters of the probability distribution of the state of the respective vehicle; The states of multiple vehicles, defined by parameters of corresponding probability distributions, are merged into an enhanced state; The measurement results of the satellite signals of the multiple vehicles are fused into an enhanced measurement result that has undergone enhanced measurement noise defined by the off-diagonal covariance matrix; A probabilistic filter is executed, which updates the enhancement state based on the enhancement measurement results subjected to the enhancement measurement noise; The parameters of the probability distribution of the vehicle state are fused with a portion of the updated enhanced state corresponding to the vehicle. as well as The parameters of the probability distribution of the fused vehicle states are sent to the RF module of the vehicle.

20. A probabilistic method that uses the asynchronous cooperation of information received from satellite signals transmitted from a Global Navigation Satellite System (GNSS) and information transmitted via a radio frequency (RF) channel to track vehicle status, wherein, The method uses a processor coupled to a memory storing a probabilistic multi-head measurement model that associates incoming measurement results with the vehicle state. The probabilistic multi-head measurement model includes a first head and a second head. The first head associates the measurement results of the satellite signal subjected to measurement noise with a confidence level of the vehicle state. The second head associates an estimate of the vehicle state subjected to estimation noise with a confidence level of the vehicle state. The processor is coupled to stored instructions implementing the method. When the instructions are executed by the processor, the steps of the method are performed, including: The measurement results of the satellite signal and the measurement noise are determined using a GNSS receiver; Receive data from the vehicle's RF receiver indicating an estimate of the vehicle's state and the estimated noise; and The parameters of the probability distribution of the vehicle state are recursively updated to generate a confidence level of the vehicle state based on the probabilistic multi-head measurement model, which accepts one or a combination of the measurement results of the satellite signal subjected to the measurement noise and the estimate of the vehicle state subjected to the estimation noise.

Images