System and method for predictive navigation control

Abstract

Systems and methods for predictive navigation control. A method for predictive navigation control of an autonomous vehicle includes: comparing a prompt node to each of a plurality of contextual memory nodes in a contextual memory structure, wherein the prompt node represents a new event, the event representing a distance, speed, and heading of one or more newly observed objects surrounding the autonomous vehicle, and wherein the contextual memory structure includes a network of nodes, each node representing a corresponding pre-existing event and having a corresponding node risk and likelihood; determining which node has a minimum corresponding difference metric, thereby defining a best matching node; merging the prompt node with the best matching node if the minimum difference metric is less than a matching tolerance, otherwise adding a new node corresponding to the prompt node to the contextual memory structure; and identifying a most likely next node and / or a most dangerous next node.

Description

Background Art

[0001] The present disclosure generally relates to systems and methods for predictive navigation control of autonomous and semi-autonomous vehicles.

[0002] Autonomous vehicles and semi-autonomous vehicles (sometimes referred to as autonomous vehicles or host vehicles) provide a degree of self-navigation (i.e., navigation of the vehicle without real-time human intervention). This self-navigation may rely on the vehicle's control system to have real-time knowledge of the surrounding environment and the vehicle's position within that environment. To accomplish this, the vehicle's control system may use a combination of stored information (e.g., maps), information received periodically or in real-time from outside the vehicle via sensors, and logic programmed into the control system. Summary of the invention

[0003] According to one embodiment, a method for predictive navigation control of an autonomous vehicle includes: (a) comparing a cue node to each of a plurality of contextual memory nodes in an episodic memory structure, wherein the cue node represents a new event that is associated with a set of corresponding positions, velocities, and headings associated with one or more newly observed objects, wherein each object is located within a corresponding one of a plurality of newly defined regions of interest surrounding the autonomous vehicle, and wherein the contextual memory structure includes a selectively interconnected and directed network of the contextual memory nodes, wherein each of the contextual memory nodes represents a corresponding pre-existing event, wherein each of the pre-existing events is associated with a corresponding set of positions, velocities, and headings associated with one or more previously observed objects, wherein each object is located within a corresponding one of a plurality of previously defined regions of interest surrounding the autonomous vehicle, and wherein the contextual memory structure includes a selectively interconnected and directed network of the contextual memory nodes, wherein each of the contextual memory nodes represents a corresponding pre-existing event, wherein each of the pre-existing events is associated with a corresponding set of positions, velocities, and headings associated with one or more previously observed objects, wherein each object is located within a corresponding one of a plurality of previously defined regions of interest surrounding the autonomous vehicle. (a) determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding episodic memory node and one or more corresponding characteristics of the prompt node; (c) merging the prompt node with the best matching node if the minimum corresponding difference metric is less than a predetermined matching tolerance, or adding a new episodic memory node corresponding to the prompt node to the episodic memory structure if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance; and (d) identifying one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most dangerous next episodic memory node among the one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most dangerous next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

[0004] The corresponding node risk of each contextual memory node is the corresponding maximum, average, or sum of the corresponding object risks of one or more previously observed objects in the associated pre-existing event, and the corresponding object risk for each previously observed object is determined by a sigmoidal function applied to the corresponding distance between the previously observed object and the autonomous vehicle.

[0005] The method may further include: calculating a respective individual risk for each of the one or more newly observed objects using a sigmoidal function applied to a respective distance between each respective newly observed object and the autonomous vehicle; establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed objects located within each respective newly defined region of interest; and defining prompt nodes as groups of the plurality of newly defined regions of interest organized according to their respective overall risks.

[0006] The corresponding object risk or individual risk for each previously observed or newly observed object is given by Determine, where R is the corresponding object risk or individual risk, Mindist is the distance from the ego vehicle to a previously observed or newly observed object, and where SAFEDIST is a distance that depends on the road surface type, road structure type, weather / environmental conditions, and one or more of relative lane position, closing speed, or closing acceleration between a previously observed or newly observed object and the autonomous vehicle.

[0007] Each of the respective positions, velocities, and headings of the one or more newly observed objects and the one or more previously observed objects can be defined relative to the autonomous vehicle. Each respective aggregate difference is a respective sum of one or more weighted penalties assigned to each of the one or more differences between respective characteristics of the respective contextual memory node and the cue node, and the characteristics can include the number and type of regions of interest, the number of objects in each region of interest, the type of road surface, the type of road structure, the type of environment, weather / environmental conditions, driving goals, driving modes, vehicle types, powertrain types, and respective positions, velocities, headings, distances from the autonomous vehicle and object risks associated with the one or more previously observed and / or newly observed objects.

[0008] The method may further include back-propagating a corresponding node risk associated with the high-risk contextual memory node to one or more contextual memory nodes upstream of the high-risk contextual memory node, wherein the associated node risk is greater than a predetermined risk threshold. Back-propagation of the associated node risk may utilize a linear function or a logistic function.

[0009] When a new episodic memory node is added to the episodic memory structure, the new episodic memory node is added as a child node to one or more parent nodes, where each parent node is a previously existing episodic memory node, and where the new episodic memory node is assigned an initial likelihood value. Each episodic memory node also has a corresponding node reward.

[0010] According to another embodiment, a method for optimizing decisions for a decision maker includes a comparing step, a determining step, a merging step, and an identifying step. The comparing step includes comparing a cue node with each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event associated with a set of corresponding attributes associated with one or more newly observed stimuli, each stimulus assigned to a corresponding one of a plurality of newly defined attention zones defined relative to a current state of the decision maker, and wherein the episodic memory structure includes a selectively interconnected and directed network of episodic memory nodes, wherein each episodic memory node represents a corresponding pre-existing event, wherein each pre-existing event is associated with a corresponding set of attributes associated with one or more previously observed stimuli, wherein each stimulus assigned to a corresponding one of a plurality of previously defined attention zones defined relative to a corresponding previous state of the decision maker, and wherein each episodic memory node has a corresponding node risk and a corresponding likelihood associated therewith.

[0011] The determining step includes determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding episodic memory node and one or more corresponding aspects of the prompt node. The merging step includes merging the prompt node with the best matching episodic memory node if the minimum corresponding difference metric is less than a predetermined matching tolerance, or adding a new episodic memory node corresponding to the prompt node to the episodic memory structure if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance. And the identifying step includes identifying one or both of: (i) the most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) the most dangerous next episodic memory node among the one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most dangerous next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

[0012] The corresponding node risk of each episodic memory node can be a corresponding maximum, average, or sum of corresponding stimulus risks of one or more previously observed stimuli in the associated pre-existing event, and the corresponding stimulus risk for each previously observed stimulus can be determined by a sigmoidal function applied to the corresponding influence potential between the previously observed stimuli and one or more goals of the decision maker. Each corresponding aggregate difference is a corresponding sum of one or more weighted penalties assigned to each of the one or more differences between the corresponding aspects of the corresponding episodic memory node and the cue node.

[0013] The method may further include: calculating a corresponding individual risk for each of one or more newly observed stimuli using a sigmoidal function of the corresponding influence potential between each corresponding newly observed stimulus and one or more targets of the decision maker; establishing a corresponding overall risk for each of a plurality of newly defined areas of interest based on the corresponding individual risks of the one or more newly observed stimuli assigned to each corresponding newly defined area of ​​interest; and defining prompt nodes as groups of the plurality of newly defined areas of interest organized according to their corresponding overall risks.

[0014] Each of the corresponding attributes of one or more newly observed stimuli and one or more previously observed stimuli is defined relative to one or more goals of the decision maker. The attributes may include one or more of importance, urgency, risk, reward, cost, duration, and difficulty. Aspects may include one or more of user temperament, user skill level, environmental factors, previous user habits, and corresponding importance, urgency, risk, reward, cost, duration, and difficulty associated with one or more previously observed or newly observed stimuli. And the goals of the decision maker include minimizing risk, optimizing utility, optimizing risk and reward, learning to anticipate the needs of the user, finding the most useful information, and learning how to recognize when to provide help to the user.

[0015] According to yet another embodiment, a controller for predictive navigation control of an autonomous vehicle includes a data ingestion, a context memory module, and a prediction module. The data ingestion module is configured to receive input associated with a prompt node, wherein the prompt node represents a new event associated with a set of corresponding positions, velocities, and headings associated with one or more newly observed objects, each of which is located within a corresponding one of a plurality of newly defined regions of interest surrounding the autonomous vehicle. The data ingestion module is further configured to: (i) calculate a corresponding individual risk for each of the one or more newly observed objects using a sigmoidal function applied to a corresponding distance between each corresponding newly observed object and the autonomous vehicle; (ii) establish a corresponding overall risk for each of the plurality of newly defined regions of interest based on the corresponding individual risks of the one or more newly observed objects located within each corresponding newly defined region of interest; and (iii) define the prompt node as a grouping of the plurality of newly defined regions of interest organized according to their corresponding overall risks.

[0016] The episodic memory module is operably connected to the data ingestion module and includes an episodic memory structure, the episodic memory structure comprising a selectively interconnected and directed network of episodic memory nodes, wherein each episodic memory node represents a respective pre-existing event, wherein each pre-existing event is associated with a respective set of positions, velocities, and headings associated with one or more previously observed objects, wherein each previously observed object is located within a respective one of a plurality of previously defined regions of interest surrounding the autonomous vehicle, and wherein each episodic memory node has a respective node risk and a respective likelihood associated therewith. The episodic memory module is configured to: (x) compare the prompt node with each of the episodic memory nodes; (y) determine which of the episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a respective aggregate difference between one or more corresponding characteristics of the respective episodic memory node and the prompt node; and (z) merge the prompt node with the best matching episodic memory node if the minimum difference metric is less than a matching tolerance, or add a new episodic memory node corresponding to the prompt node to the episodic memory structure if the minimum corresponding difference metric is greater than or equal to a predetermined matching tolerance.

[0017] The prediction module is operably connected to the episodic memory module and configured to identify one or both of the following: a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of a best match or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes; and a most dangerous next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most dangerous next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

[0018] The controller may further include a perception module operably connected to the data ingestion module and configured to: detect, as sensed data, a set of corresponding positions, velocities, and headings associated with one or more newly observed objects; and convert the sensed data into input associated with a prompt node for receipt by the data ingestion module.

[0019] The corresponding node risk of each contextual memory node is a corresponding maximum, average, or sum of corresponding object risks of one or more previously observed objects in the associated previously existing event, and wherein the corresponding object risk for each previously observed object is determined by a sigmoidal function applied to a corresponding distance between the previously observed object and the autonomous vehicle. The contextual memory module may be further configured to back-propagate the corresponding node risk associated with a high-risk contextual memory node to one or more contextual memory nodes upstream of the high-risk contextual memory node, wherein the associated node risk is greater than a predetermined risk threshold.

[0020] The present invention also provides the following technical solutions:

[0021] 1. A method for predictive navigation control of an autonomous vehicle, the method comprising:

[0022] comparing the cue node to each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event, the new event being associated with a corresponding set of positions, velocities, and headings associated with one or more newly observed objects, wherein each object is located within a corresponding one of a plurality of newly defined zones of interest surrounding the autonomous vehicle, and wherein the episodic memory structure comprises a selectively interconnected and directed network of the episodic memory nodes, wherein each episodic memory node represents a corresponding pre-existing event, wherein each pre-existing event is associated with a corresponding set of positions, velocities, and headings associated with one or more previously observed objects, wherein each object is located within a corresponding one of a plurality of previously defined zones of interest surrounding the autonomous vehicle, and wherein each episodic memory node has a corresponding node risk and a corresponding likelihood associated therewith;

[0023] determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between one or more corresponding characteristics of the corresponding episodic memory node and the cue node;

[0024] If the minimum corresponding difference measure is less than a predetermined matching tolerance, merging the hint node with the best matching node, or if the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance, adding a new episodic memory node corresponding to the hint node to the episodic memory structure; and

[0025] Identifying one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most risky next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

[0026] 2. A method according to technical solution 1, wherein the corresponding node risk of each situational memory node is the corresponding maximum, average or sum of the corresponding object risks of the one or more previously observed objects in the associated previously existing events, and wherein the corresponding object risk for each previously observed object is determined by an inverse function applied to the corresponding distance between the previously observed object and the autonomous vehicle.

[0027] 3. The method according to technical solution 2 further comprises:

[0028] calculating a respective individual risk for each of the one or more newly observed objects using the sigmoidal function applied to a respective distance between each respective newly observed object and the autonomous vehicle;

[0029] establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed objects located within each respective newly defined region of interest; and

[0030] The prompt nodes are defined as groupings of the plurality of newly defined areas of interest organized according to their respective overall risks.

[0031] 4. A method according to claim 3, wherein the corresponding object risk or individual risk for each previously observed or newly observed object is determined by Determine, where R is the corresponding object risk or individual risk, Mindist is the distance from the autonomous vehicle to the previously observed or newly observed object, and where SAFEDIST is a distance that depends on road surface type, road structure type, weather / environmental conditions, and one or more of relative lane position, closing speed, or closing acceleration between the previously observed or newly observed object and the autonomous vehicle.

[0032] 5. A method according to technical solution 1, wherein each of the corresponding positions, velocities and directions of travel of the one or more newly observed objects and the one or more previously observed objects is defined relative to the autonomous vehicle.

[0033] 6. A method according to technical solution 1, wherein each corresponding aggregate difference is a corresponding sum of one or more weighted penalties assigned to each of one or more differences between the corresponding characteristics of the corresponding contextual memory node and the prompt node, and wherein the characteristics include the number and type of focus areas, the number of objects in each focus area, road surface type, road structure type, environment type, weather / environmental conditions, driving goals, driving modes, vehicle type, powertrain type, and the corresponding position, speed, heading, and distance of the autonomous vehicle and object risks associated with the one or more previously observed and / or newly observed objects.

[0034] 7. The method according to technical solution 1 further comprises:

[0035] Back-propagating a corresponding node risk associated with a high-risk episodic memory node to one or more episodic memory nodes upstream of the high-risk episodic memory node, wherein the associated node risk is greater than a predetermined risk threshold.

[0036] 8. A method according to Technical Solution 7, wherein the back propagation of the associated node risks utilizes a linear function or a logical function.

[0037] 9. A method according to technical solution 1, wherein when a new situational memory node is added to the situational memory structure, the new situational memory node is added as a child node to one or more parent nodes, wherein each parent node is a previously existing situational memory node, and wherein the new situational memory node is assigned an initial likelihood value.

[0038] 10. The method according to technical solution 1, wherein each contextual memory node also has a corresponding node reward.

[0039] 11. A method for optimizing decision making for a decision maker, the method comprising:

[0040] comparing the cue node to each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event associated with a corresponding set of attributes associated with one or more newly observed stimuli, each stimulus assigned to a corresponding one of a plurality of newly defined attention regions defined relative to a current state of the decision maker, and wherein the episodic memory structure comprises a selectively interconnected and directed network of the episodic memory nodes, wherein each episodic memory node represents a corresponding pre-existing event, wherein each pre-existing event is associated with a corresponding set of attributes associated with one or more previously observed stimuli, wherein each stimulus assigned to a corresponding one of a plurality of previously defined attention regions defined relative to a corresponding previous state of the decision maker, and wherein each episodic memory node has a corresponding node risk and a corresponding likelihood associated therewith;

[0041] determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between one or more aggregate aspects of the corresponding episodic memory node and the cue node;

[0042] If the minimum corresponding difference metric is less than a predetermined matching tolerance, merging the hint node with the best matching node, or if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance, adding a new episodic memory node corresponding to the hint node to the episodic memory structure; and

[0043] Identifying one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best matching or new episodic memory node, wherein the most risky next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

[0044] 12. The method according to technical solution 11, wherein each of the corresponding attributes of the one or more newly observed stimuli and the one or more previously observed stimuli is defined relative to one or more objectives of the decision maker.

[0045] 13. A method according to technical solution 12, wherein the corresponding node risk of each situational memory node is the corresponding maximum, average or sum of the corresponding stimulus risks of the one or more previously observed stimuli in the associated previously existing events, and wherein the corresponding stimulus risk for each previously observed stimulus is determined by an inverse function of the corresponding influence potential between the previously observed stimuli and the one or more targets of the decision maker.

[0046] 14. The method according to technical solution 13 further comprises:

[0047] calculating a respective individual risk for each of the one or more newly observed stimuli using the sigmoidal function applied to the respective impact potential between each respective newly observed stimulus and the one or more objectives of the decision maker;

[0048] establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed stimuli assigned to each respective newly defined region of interest; and

[0049] The prompt nodes are defined as groupings of the plurality of newly defined areas of interest organized according to their respective overall risks.

[0050] 15. A method according to technical solution 11, wherein each corresponding aggregate difference is a corresponding sum of one or more weighted penalties assigned to each of one or more differences between the corresponding aspects of the corresponding contextual memory node and the prompt node.

[0051] 16. A method according to technical solution 12, wherein: (i) the attributes include one or more of importance, urgency, risk, reward, cost, duration and difficulty; (ii) the aspects include user temperament, user skill level, environmental factors, previous user habits and one or more of the corresponding importance, urgency, risk, reward, cost, duration and difficulty associated with the one or more previously observed or newly observed stimuli; and (iii) the goals of the decision maker include minimizing risk, optimizing utility, optimizing risk and reward, learning to anticipate user needs, finding the most useful information and learning how to identify when to provide help to the user.

[0052] 17. A controller for predictive navigation control of an autonomous vehicle, the controller comprising:

[0053] (a) a data ingestion module configured to receive input associated with a hint node, wherein the hint node represents a new event associated with a set of respective positions, velocities, and headings associated with one or more newly observed objects, each of the newly observed objects being located within a respective one of a plurality of newly defined regions of interest surrounding the autonomous vehicle, the data ingestion module being further configured to:

[0054] calculating a respective individual risk for each of the one or more newly observed objects using a sigmoidal function applied to a respective distance between each respective newly observed object and the autonomous vehicle;

[0055] establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed objects located within each respective newly defined region of interest; and

[0056] defining the alert nodes as a grouping of the plurality of newly defined areas of interest organized according to their respective overall risks;

[0057] (b) an episodic memory module operably connected to the data ingestion module and comprising an episodic memory structure, the episodic memory structure comprising a selectively interconnected and directed network of episodic memory nodes, wherein each episodic memory node represents a respective pre-existing event, wherein each pre-existing event is associated with a respective set of positions, velocities, and headings associated with one or more previously observed objects, wherein each previously observed object is located within a respective one of a plurality of previously defined regions of interest surrounding the autonomous vehicle, and wherein each episodic memory node has a respective node risk and a respective likelihood associated therewith, the episodic memory module being configured to:

[0058] comparing the prompt node with each of the episodic memory nodes;

[0059] determining which of the episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between one or more corresponding characteristics of the corresponding episodic memory node and the cue node; and

[0060] merging the hint node with the best matching node if the minimum difference metric is less than a matching tolerance, or adding a new episodic memory node corresponding to the hint node to the episodic memory structure if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance; and

[0061] (c) a prediction module operably connected to the episodic memory module and configured to identify one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most likely next episodic memory node has a highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most risky next episodic memory node has a highest node risk among the immediately downstream episodic memory nodes.

[0062] 18. The controller according to technical solution 17 further comprises:

[0063] (d) a perception module, the perception module being operatively connected to the data ingestion module and configured to:

[0064] detecting, as sensed data, the set of corresponding positions, velocities, and headings associated with the one or more newly observed objects; and

[0065] The sensed data is converted into the input associated with the prompt node for receipt by the data ingestion module.

[0066] 19. A controller according to technical solution 17, wherein the corresponding node risk of each context memory node is the corresponding maximum, average or sum of the corresponding object risks of the one or more previously observed objects in the associated previously existing events, and wherein the corresponding object risk for each previously observed object is determined by the inverse function applied to the corresponding distance between the previously observed object and the autonomous vehicle.

[0067] 20. The controller according to technical solution 17, wherein the context memory module is further configured:

[0068] Back-propagating a corresponding node risk associated with a high-risk episodic memory node to one or more episodic memory nodes upstream of the high-risk episodic memory node, wherein the associated node risk is greater than a predetermined risk threshold.

[0069] The above features and advantages and other features and advantages of the present teachings as defined in the appended claims are readily apparent from the following detailed description of some of the best modes and other embodiments for carrying out the present teachings when considered in conjunction with the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0070] Figure 1is a perspective overhead view of the autonomous vehicle and other vehicles on a five-lane highway representing a new event.

[0071] Figure 2 is a graph showing the selected area of ​​interest Figure 1 top view of vehicles and highway.

[0072] Figure 3 is a diagram of a method for predictive navigational control showing comparison of cue nodes with episodic memory nodes using a common event structure.

[0073] Figure 4 is a diagram showing the cue node and episodic memory structure.

[0074] Figure 5 is a perspective overhead view of the autonomous vehicle and other vehicles on a three-lane highway showing a pre-existing event.

[0075] Figure 6 is a graph showing the selected area of ​​interest Figure 5 top view of vehicles and highway.

[0076] Figure 7 is a flow chart of a method for predictive navigation control.

[0077] Figure 8 is a block diagram of a system / controller for implementing a method of predictive navigation control.

[0078] Fig. 9 is a block diagram of a system / controller for implementing a method for optimizing decisions for a decision maker.

[0079] Fig.10 is a flow chart of a method for optimizing decisions for a decision maker.

[0080] Fig.11 is a diagram of the decider and various stimuli.

[0081] Fig.12 is a graph showing the selected area of ​​interest Fig.11 Illustration of the decision maker and stimulus.

[0082] Fig.13 is a diagram showing a method for optimizing decisions for a decision maker comparing a cue node to an episodic memory node using a common event structure.

[0083] Figures 14 to 16 A block diagram of the various characteristics, aspects and attributes, and goals, respectively. DETAILED DESCRIPTION

[0084] Referring now to the drawings, wherein like numerals indicate like parts throughout the several views, a method 100 for predictive navigation control of an autonomous vehicle 20, a method 300 for optimizing decisions for a decision maker 50, and a system / controller 230, 430 for implementing the respective methods 100, 300 are shown and described herein. Note that certain reference numerals in the drawings have subscripts, such as Figures 2 to 3 Three areas of concern32 L , 32 H and 32 U Subscripts are used in the drawings and this specification to refer to individual elements (such as the aforementioned region of interest), while the use of reference numerals without subscripts may refer to the aggregate group of such elements and / or a singular but common one of such elements. Thus, reference numeral 32 L refers to a specific region of interest, whereas reference numeral 32 (without a subscript) may refer to all regions of interest, a group of regions of interest, or a singular but generic region of interest (ie, any region of interest).

[0085] It should be noted that although Figure 1 to Figure 2 and Figures 5 and 6 An exemplary traffic situation involving automotive passenger vehicles is shown, but the methods 100, 300 and systems / controllers 230, 430 of the present disclosure are also applicable to other types of autonomous and semi-autonomous vehicles and environments, such as drones and aircraft flying in airspace, ground vehicles traveling on roads or terrain, mobile machines moving on the ground and / or through structures, water vessels moving on the surface of a body of water, submarines moving on the surface of a body of water or at different depths below the body of water, satellites or spacecraft moving through space, etc. It is also noted that while straight five-lane or three-lane highway segments are used in the drawings and specifications, the methods 100, 300 and systems / controllers 230, 430 are also applicable to other road and lane configurations, including intersections, merging / diverging lanes, curves, parking lots, open terrain, etc.

[0086] Figures 1 to 6 The following process is shown: Take a given event 30 such as a traffic situation involving an autonomous vehicle 20 ( Figure 1 ); Identify the areas of interest 32 around the autonomous vehicle 20 and the arrangement of other neighboring vehicles / objects placed in those areas of interest 32 ( Figure 2 ); Organize the collection of areas of interest 32 and other data according to the event structure 39 ( Figure 3 ); and representing this information collection as a cue node 34 that can be compared to individual episodic memory nodes 36 in an episodic memory structure 38 ( Figure 4 ), where each episodic memory node 36 represents a corresponding previously existing event 40 ( Figures 5 and 6). (Hereinafter, the abbreviation "EM" will be used to represent the word "episodic memory." Note also that "around" autonomous vehicle 20 means around, in front of, behind, beside, and / or adjacent to autonomous vehicle 20, and that "autonomous vehicle" 20 may refer to (i) the current host vehicle that is experiencing new event 30 that is being converted to prompt node 34, (ii) the current host vehicle that has previously experienced event 40 that has been converted to EM node 36, and / or (iii) other host vehicles that have experienced previous events 40 that have been converted to EM nodes 36 that have been imported into the current host vehicle.) If a prompt node 34 sufficiently matches one of EM nodes 36 in EM structure 38, then prompt node 34 may be merged or combined into a matching EM node 36; however, if a prompt node 34 does not sufficiently match any of EM nodes 36, then prompt node 34 may be added to EM structure 38 as a new EM node 36. This process will now be described in more detail below.

[0087] Figure 1 A perspective top view of an autonomous vehicle 20 and other vehicles 21, 22, 23, 24, 25 on a five-lane highway is shown. Autonomous vehicle 20 is shown cruising in fourth lane 27 from the left, and another vehicle 21 (sometimes referred to herein as "neighboring vehicle 21") is shown positioned adjacent to the right rear of autonomous vehicle 20 in fifth lane 28 from the left. Another vehicle 22 (sometimes referred to herein as "leading vehicle 22") is shown directly in front of autonomous vehicle 20 in the same fourth lane 27 as autonomous vehicle 20. This aggregate of vehicles 21, 22, 23, 24, 25 surrounding autonomous vehicle 20, their respective positions L, speeds S, and headings H relative to autonomous vehicle 20, and the like are shown in FIG. Figure 1 All of these together provide a “snapshot” in time of a particular driving situation or event, which may be referred to as a new event 30 .

[0088] Figure 2 A schematic top view of autonomous vehicle 20, neighboring vehicle 21, and lead vehicle 22 within a selected region of interest 32 is shown. Vehicles / objects outside of region of interest 32, such as vehicle 23 (in lane 26) and vehicles 24 and 25, are not shown or considered in this step. (Note that as used herein, "vehicle," "object," and "vehicle / object" are used interchangeably. The plural forms of these terms are also used interchangeably.) In the context of autonomous vehicle 20 on a highway, region of interest 32 is a section of road having a predetermined shape, size, and position relative to autonomous vehicle 20. For example, as shown in FIG. Figure 2As shown in , each region of interest 32 may be a generally rectangular space that is a lane that is about three car lengths wide and long. The size and shape of the region of interest 32 may be fixed, or the size and shape may vary depending on factors such as the speed of the autonomous vehicle, the type of road surface or terrain, weather conditions, the curvature of the lane, etc. Also note that in Figure 2 In the example shown, only some portions of the highway have been indicated as regions of interest 32; specifically, there is a region of interest 32 in the center of the autonomous vehicle 20. e , there is a region of interest 32 on either side of the autonomous vehicle 20 l , 32 r There are three interest areas 32 in a row directly behind the autonomous vehicle 20. b1 , 32 b2 , 32 b3 , and there are two rows of three interest areas 32 directly in front of the autonomous vehicle 20 a1 , 32 a2 , 32 a3 , 32 a4 , 32 a5 , 32 a6 Note that any number and arrangement of regions of interest 32 may be used. Note also that the arrangement of regions of interest 32 surrounding autonomous vehicle 20 will dynamically move along with autonomous vehicle 20, and that the arrangement of regions of interest 32 may optionally change as autonomous vehicle 20 changes speed, changes direction of travel, moves to a different lane, moves to a different road surface, moves to a different road condition, experiences changing weather conditions, etc.

[0089] The areas of interest 32 may optionally be divided into two or more types, such as low risk areas 32 L 、High risk area 32 H and emergency risk zone 32 U For example, the rightmost region of interest 32 of the autonomous vehicle 20 r This is shown by widely spaced hatching, which demarcates this region of interest 32 r Identified as a high risk area32 H (Due to the close proximity of the neighboring vehicle 22). The region of interest 32 directly in front of the ego vehicle 20 a5 This is shown by widely spaced hatching, which demarcates this region of interest 32 a5 Identified as an urgent risk area32 U (Due to the potential for the lead vehicle 22 to brake and thus interfere with the travel of the ego vehicle 20). Autonomous vehicle's area of ​​interest 32 e and nine other areas of concern32 l , 32 bl , 32 b2 , 32 b3 , 32 a1 , 32 a2, 32 a3 , 32 a4 , 32 a6 Shown in light grey shading, this identifies these as low risk areas 32 L .

[0090] Based on various risk factors, such as the respective position L, speed S, and heading H of one or more vehicles / objects in the area of ​​interest 32, the area of ​​interest 32 may be characterized as a low risk area 32. L 、High risk area 32 H or emergency risk zone 32 U These characteristics of position L, speed S, and heading H can be considered to affect the proximity of each vehicle / object to the autonomous vehicle 20 and the potential of each vehicle / object to interfere with or adversely affect the forward progress of the autonomous vehicle 20. For example, the distance d from the autonomous vehicle 20 to the neighboring vehicle 21 is 21 and the location of the neighboring vehicles 21 to the side and rear of the ego vehicle 20 may result in the area of ​​interest 32 b3 Characterized as high risk area32 H , and the distance d from the autonomous vehicle 20 to the leading vehicle 22 22 and the position of the lead vehicle 22 directly in front of the ego vehicle 20 may result in the region of interest 32 a5 Characterized as emergency risk area 32 U .

[0091] Once regions of interest 32 surrounding autonomous vehicle 20 have been identified, and each region of interest 32 has been optionally categorized as a low risk region 32L, a high risk region 32H, or an urgent risk region 32U, the respective position L, speed S, and heading H of the vehicle / object within each region of interest 32 may be determined relative to the position, speed, and heading of autonomous vehicle 20, and these relative positions L, relative speeds S, and relative headings H may be stored in a format such as Figure 3 The event structure 39 illustrated in FIG. The event structure 39 may be a data structure, ordered data set, table, organized grouping, format, etc., in which the characteristics 49 of the position L, speed S and heading H for each vehicle / object are stored (and other optional data such as distance d and individual risk IR).

[0092] like Figure 3 As shown in , event 30 (e.g., Figure 1 to Figure 2 traffic conditions) and three areas of concern32 - a low risk area32 L , a high risk area 32 H and an emergency risk zone 32 U (Note that for simplicity, Figure 8 Only one low risk area 32 is shown L , but if Figure 2 As shown in FIG. , nine low risk areas 32 may be shown. L . ) Elements associated with the low risk area 32L are drawn with dashed lines to indicate that the low risk area 32 L No vehicle / object exists in High Risk Area 32 H is shown with position L 21 , speed S 21 and the forward direction H 21 A vehicle 21 and the emergency risk area 32U are also shown as having a position L 22 , speed S 22 and the forward direction H 22 A vehicle 22. Note that although Figure 3 2 and 3. The neighboring vehicles 21 and the leading vehicle 22 are shown in FIG. 3 (each represented as a hexagon), but the event structure 39 may optionally exclude the identification of these individual vehicles 21, 22 and may instead include the corresponding positions L of these vehicles 21, 22. 21 , L 22 , speed S 21 , S 22 and the forward direction H 21 , H 22 and the corresponding distance d 21 d 22 and individual risk IR 21 IR 22 And other relevant information.

[0093] The event structure 39 may also include a distance d and an individual risk IR for each vehicle / object, where the distance d represents the distance between the vehicle / object and the autonomous vehicle 20, and the individual risk IR for each vehicle / object may be determined by a sigmoidal function applied to the distance d. (e.g., as Figure 2 As shown in 21 is the distance between the neighboring vehicle 21 and the ego vehicle 20, and d 22 is the distance between the lead vehicle 22 and the ego vehicle 20, and the corresponding individual risk IR 21 IR 22 These distances d can be used 21 d 22 and when the inverse equation is determined, an example of which is provided below.)

[0094] like Fig.14 As shown in FIG. 4 , the characteristics 49 stored in the event structure 39 may also include additional information AI, such as the number and type of the regions of interest 32 (AI AZ ) (For example, Figure 2 and Figure 5 and utilizing the low, high and emergency zones 32 discussed above.L , 42 L , 32 H , 42 H , 32 U , 42 U A grid of twelve regions 42 of similar three types of regions 42); the number of vehicles / objects in each region of interest 32 (AI NO ); Road surface type (AI RS ) (e.g., paved, gravel, dirt, etc.); road structure (AI RT ) (e.g., number of lanes, lane directions, etc.); environment type (AI E ) (e.g., city, residential, highway, parking lot, etc.); weather / environmental conditions (AI WC ) (e.g., sunny, light rain, heavy rain, snow, etc.); driving goal (AI DG ) (e.g., fastest route, leisure / scenic route, etc.); driving mode (AI DM ) (e.g., cruise, sports, fuel economy, etc.); vehicle type (AI VT ) (e.g., passenger car, truck, gross vehicle weight, horsepower, etc.); and powertrain type (AI PT ) (e.g., automatic, manual, electric, hybrid, number of gears, etc.). For example, Figure 3 As shown in FIG. 21 One or more additional information AI related to the neighboring vehicle 21 may be represented. This organized collection of information stored in the format of the event structure 39 may be used as a digital representation of the event 30 and as Figure 3 As shown in , it can be represented as a prompt node 34. Note that Figure 3 It is also shown that the event structure 39 of the prompt node 34 and its associated new event 30 also apply to the EM node 36 and its associated pre-existing event 40, and the additional information AI also applies to both events 30, 40 (i.e., AI 30 、AI 40 ).

[0095] A system or process utilizing the methods described herein may begin by building an EM structure 38 from scratch, or by receiving or importing a previously created EM structure 38, which may have been created by another autonomous vehicle that has similarities (i.e., similar vehicle type, similar powertrain, etc.) to the present autonomous vehicle 20. In either case, the system and method may follow the following process: sensing or receiving information about each new event 30 ( Figure 1 ); Identify the arrangement of the regions of interest 32 around the autonomous vehicle 20 and other neighboring vehicles / objects placed within those regions of interest 32 ( Figure 2); organize information about the region of interest 32 and the vehicles / objects therein in the format of an event structure 39 ( Figure 3 ); and converting the organized information into hint nodes 34 that can numerically represent the new event 30. The hint nodes 34 can then be compared to individual EM nodes 36 in an EM structure 38, where each EM node 36 represents a previously existing event 40. Whether the system and method builds the EM structure 38 from scratch or starts with an imported EM structure 38, the system and method can continue to expand the EM structure 38 over time as new events 30 are encountered, converting the new events 30 into hint nodes 36, and comparing the new events 30 to nodes 36 in an existing EM structure 38.

[0096] like Figure 3 As shown in FIG. 1 , such comparison between the prompt node 34 and each EM node 36 can be facilitated if each EM node 36 utilizes the same event structure 39 as the prompt node 34. Here, the exemplary EM node 36 represents an event structure 39 related to a prompt node 34. Figure 4 The previously observed vehicle / object shown in FIG. 4 corresponds to a previously existing event 40 . Figure 4 In FIG. 4 , the autonomous vehicle 20 is traveling in the middle lane 47 of a three-lane highway. One neighboring vehicle 43 is beside the autonomous vehicle 20 in the right lane 48, and two leading vehicles 44, 45 are in front of the autonomous vehicle 20 in the middle lane 47. No vehicles / objects are shown in the left lane 46. In terms of the overall positions of other neighboring vehicles / objects around the autonomous vehicle 20, this traffic situation is similar to Figure 1 situation.

[0097] Figure 5 shows a diagram with an identified region of interest 42 Figure 4 Schematic diagram of event 40 shown in FIG. Note that vehicle 45 is not included in Figure 5 The area of ​​interest 42 that includes the neighboring vehicle 43 (i.e., the area of ​​interest 42 r ) are shown with widely spaced hatching to indicate this as a high risk area 42 H , and includes the area of ​​interest 42 of the leading vehicle 44 (ie, the area of ​​interest 42 a5 ) is shown with closely spaced hatching to indicate it as an emergency risk area 42U. In addition to the area of ​​interest 42 (i.e., the area of ​​interest 42) containing the autonomous vehicle 20 e ), the other nine areas of interest 42 (i.e. 42 l , 42 b1 , 42 b2 , 42 b3 , 42 a1 , 42 a2 , 42 a3 , 42a4 and 42 a6 ) are shown in light grey shading to indicate them as low risk areas 42 L Even if the distance d from the autonomous vehicle 20 to the leading vehicle 44 44 ( Figure 5 ) is greater than the distance d from the autonomous vehicle 20 to the leading vehicle 22 22 ( Figure 2 ) is slightly longer, and the distance d43 to the adjacent vehicle 43 is slightly shorter than the distance d21 to the adjacent vehicle 21, however Figure 2 and Figure 5 The region of interest illustrations are often very similar to each other.

[0098] Each of the two previously observed vehicles 43, 44 within the region of interest 42 has a corresponding position L stored in the event structure 39 43 , L 44 , speed S 43 , S 44 and the forward direction H 43 , H 44 , and each vehicle has a corresponding distance d to the ego vehicle 20 43 d 44 and based on the corresponding distance d 43 d 44 The corresponding object risk OR 43 , OR 44 EM nodes 36 may also have a node risk NR and a likelihood LL associated therewith. The node risk NR represents the amount of risk of the associated pre-existing event 40 being caused by the travel of the autonomous vehicle 20 (e.g., due to the proximity of neighboring vehicles, etc.), and the likelihood LL represents the likelihood of the event 40 associated with the node 36 occurring (e.g., expressed as a decimal from 0.00 to 1.00 with the likelihood LL being expressed as a percentage). Each EM node 36 may optionally have a node risk NR and a likelihood LL associated therewith. Figure 3 The corresponding node reward NW is shown in dashed outline in FIG. The node reward NW may include a digital representation of one or more rewards or benefits that can be achieved if the associated event 40 occurs.

[0099] Notice, Figure 3 The exemplary EM node 36 has very similar content to the prompt node 34. That is, the event 30 represented by the prompt node 34 has a vehicle 21 in a high risk area 32. H and having a vehicle 22 in an emergency risk zone 32 U and similarly, the event 40 represented by the exemplary EM node 36 has a vehicle 43 in a high risk zone 42 H and having a vehicle 44 in an emergency risk zone 42U For the purpose of illustration, it can also be assumed that the position L of the adjacent vehicle 21 21 , speed S 21 and the forward direction H 21 The position L of the vehicle 43 43 , speed S 43 and the forward direction H 43 Very similar (so that the distance d 21 and individual risk IR 21 Respectively with the distance d 43 and object risk OR 43 is very similar); and the position L of the leading vehicle 22 22 , speed S 22 and the forward direction H 22 The position L of the vehicle 44 44 , speed S 44 and the forward direction H 44 Very similar (so that the distance d 22 and individual risk IR 22 Respectively with the distance d 44 and object risk OR 44 In addition, and as further described below, it can be assumed that Figure 3 An exemplary EM node 36 represents Figure 4 EM node 361, and in Figure 4 Of all EM nodes 36 within EM node 38, EM node 361 is the best match for prompt node 34. However, it should be noted that although Figure 3 An exemplary node 36 (and Figure 4 The content and value of EM node 361 closely matches the content and value of prompt node 34, but the content and value of other EM nodes 36 may not so closely match prompt node 34. For example, Figure 4 The EM nodes 36 in the image may be in their respective three caution / risk zones 42 L , 42 H , 42 U With different numbers of vehicles / objects in the three regions of interest, the three regions of interest have different positions L, speeds S and headings H (and therefore different distances d and different object risks OR).

[0100] like Figure 6 As shown in FIG. 1 , EM structure 38 is a selectively interconnected and directed network of EM nodes 36. (Here, each EM node 36 has been given a unique subscript from 0 to 22, i.e., 360 to 36 22. ) Each EM node 36 represents a unique pre-existing event 40, and the aggregate of EM nodes 36 represents a scenario or situation that may involve various possible sequences of one or more of these events 40. Figure 4 360, 362, 367, 368, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 381, 382, ​​383, 384, 385, 386, 387, 388, 389, 390, 400, 410, 420, 430, 441, 451, 462, 473, 480, 490, 500, 511, 521, 531, 532, 541, 553, 564, 576, 580, 590, 600, 611, 621, 631, 10 and 36 11 Each arrow indicates that once the event 40 represented by EM node 301 has occurred, the possible next event 40 represented by EM nodes 360, 362, 367, 368, 369, 370, 371, 372, 373, 374, 375, 376, 377, 378, 379, 380, 381, 382, ​​38 10 and 36 11 The direction of each arrow indicates a possible flow from an "upstream" EM node 36 to a "downstream" EM node 36, and a possible flow from a "parent" EM node 36 to a "child" EM node 36. Thus, EM node 361 is in the order of 360, 362, 367, 368, 10 and 36 11 and each EM node 360, 362, 367, 36 10 and 36 11 is downstream of node 361; similarly, EM node 361 will be downstream of each EM node 360, 362, 367, 36 10 and 36 11 The parent node of EM nodes 360, 362, 367, 36 10 and 36 11 will be a child node to EM node 361. Note that the upward arrow in the diagram can still indicate the downstream flow direction (as the arrow from EM node 36 11 points to EM node 360); therefore, the direction of the arrow and not necessarily the location of EM node 36 on the page determines the downstream / upstream flow direction from node to node.

[0101] As indicated above, the EM structure 38 represents an aggregate of possible sequences of events 40 for a given scenario or situation. For example, in the case of a motor vehicle, each scenario may be a specific stance situation, such as cruising, passing, turning off the freeway at an exit, parking, backing up, etc. Each of these stance situations or scenarios may have its own unique EM structure 38, wherein each EM structure 38 includes an aggregate of possible sequences of events 40. Thus, a system or process utilizing the methods described herein may have access to a plurality of different stored EM structures 38, which may be selectively utilized and switched among them to match the current driving situation.

[0102] Figure 3 The solid arrows at show that the hint node 34 is compared to the single EM node 36, and Figure 4 The five solid arrows in FIG. 3 show that the hint node 34 is compared with five specific EM nodes 36 in the EM structure 38. (Although Figure 4 Only five comparisons are shown in FIG. 1 , but this is merely for simplicity because in actual practice, the prompt node 34 will be compared with each of the EM nodes 36. ) Each of the EM nodes 36 may utilize an event structure 39 format to represent its corresponding region of interest 42 and various event-related information, which may facilitate comparison of the prompt node 34 with each of the EM nodes 36.

[0103] As mentioned above, if the hint node 34 sufficiently matches one or more of the EM nodes 36, then the hint node 34 may be merged or incorporated into the best matching EM node 361; however, if the hint node 34 does not sufficiently match any of the EM nodes 36, then the hint node 34 may be used as a new EM node 36. 22 Added to EM structure 38. (Note that although reference numerals 361 and 36 22 exist Figure 4 36, but these reference numerals are also used herein to generally refer to the best matching node 361 and the new node 36, respectively. 22 . ) For example, as previously described, if it is determined that the content and value of EM node 361 has the best match with the content and value of hint node 34, and if this match is close enough (e.g., less than some predetermined amount or threshold), then hint node 34 can be merged into EM node 361; however, if the match between node 361 and hint node 34 is not close enough, then hint node 34 can be used as a new node, such as node 361. 22 , added to the EM structure 38, the node 36 22 Already attached to EM node 36 20 and is shown in dashed outline. Once the hint node 34 has been merged into the best matching node 361, or as a new node 36 22 Add, assuming the best matching node or new node 361, 36 22 (depending on the situation) the most likely node and / or the most dangerous node 362, 36 10 Node.

[0104] Based on the respective aggregate differences between the one or more characteristics 49 of the EM node 36 and the same one or more characteristics 49 of the hint node 34, a respective difference metric may be determined for each EM node 36. Once the difference metrics are determined for all EM nodes 36, the EM node 36 with the smallest difference metric may be identified as the best matching EM node 361. Next, if the difference metric of the best matching EM node 361 is less than a predetermined matching tolerance, the hint node 34 is merged into the best matching node 361; otherwise, the hint node 34 is merged as the new node 361. 22 Added to EM structure 38.

[0105] When a hint node 34 is merged or incorporated into a best matching EM node 361, the merger or merging may optionally have an impact on the best matching EM node 361. For example, the merger may result in increasing or increasing the likelihood LL of the best matching EM node 361. On the other hand, when a hint node 34 is used as a new EM node 361, the hint node 34 may be merged or incorporated into the best matching EM node 361. 22 When adding, it can be added as a child node to the EM node 36 that best matches it. Figure 4 Assuming EM node 36 20 . (That is, the new EM node 36 22 can be added as a child node to one or more parent nodes, where each parent node is a previously existing EM node 36.) Note that although the new node 36 22 Shown here as having no children, but in some cases, a new node 3622 may be inserted between (and connected to) two or more EM nodes 36, in which case the new node 3622 may be a child node of the EM node 36. 22 Will have one or more child nodes.

[0106] Now, the foregoing method will be described in more detail, according to one embodiment (and as described by Figure 7 ), the method 100 for predictive navigation control of an autonomous vehicle 20 includes a comparison step (box 160), a determination step (box 170), a merging step (box 200), and an identification step (box 220). At box 160, the comparison step includes comparing a hint node 34 with each of a plurality of EM nodes 36 in an EM structure 38, wherein the hint node 34 represents a new event 30. The new event 30 is associated with a set of corresponding positions L, velocities S, and headings H, which in turn are associated with one or more newly observed objects 21. (Note that although reference numeral 21 is not shown in FIG. 1 , FIG. 1 ) Figure 1 to Figure 23 to denote a neighboring vehicle 21, but this reference numeral is also used herein to denote one or more newly observed objects 21.) Each newly observed object 21 is located within a respective one of a plurality of newly defined regions of interest 32 surrounding the autonomous vehicle 20. The EM structure 38 includes a selectively interconnected and directed network of EM nodes 36, wherein each EM node 36 represents a respective pre-existing event 40. Each pre-existing event 40 is associated with a respective set of positions L, velocities S, and headings H associated with one or more previously observed objects 43. (Note that although reference numeral 43 is used in Figures 4 to 5 4 to denote a particular vehicle / object 43, but this reference numeral is also used herein to denote one or more previously observed objects 43.) Each previously observed object 43 is located within one of a plurality of previously defined regions of interest 42 surrounding the autonomous vehicle 20. In addition, each EM node 36 has a corresponding node risk NR and a corresponding likelihood LL (and optionally a corresponding node reward NW) associated therewith.

[0107] At block 170, the determining step includes determining which of the plurality of EM nodes 36 has the smallest corresponding difference metric, thereby defining the best matching EM node 361. Each corresponding difference metric is determined based on a corresponding aggregate difference between one or more corresponding characteristics of the corresponding EM node 36 and the prompt node 34. Fig.14 As shown in FIG. 4 , these characteristics 49 may include the number and type of regions of interest 32, 42 (AI AZ ), the number of objects in each area of ​​interest 32, 42 (AI NO ), road surface type (AI RS ), road structure type (AI RT ), environment type (AI E ), weather / environmental conditions (AI WC ), driving objectives (AI DG ), driving mode (AI DM ), vehicle type (AI VT )、Powertrain Type(AI PT ) and the corresponding position L, speed S, heading H, distance d from the autonomous vehicle 20 and individual / object risks IR, OR associated with one or more previously observed and / or newly observed objects 43, 21.

[0108] Each of the respective positions L, velocities S, and headings H of the one or more newly observed objects 21 and the one or more previously observed objects 43 may be defined relative to the autonomous vehicle 20. (That is, the autonomous vehicle 20 may serve as the origin of a reference coordinate axis, wherein the positions L, velocities S, and headings H may be defined relative to the autonomous vehicle 20.) The aggregate difference may be the sum, average, and maximum of the differences between or among these characteristics 49. For example, using Figure 1 to Figure 2 Event 30 and Figure 3 The total difference can be the sum of the differences between the corresponding positions L of the vehicles / objects; that is, (L 44 –L 22 ) + (L 43 –L 21 ), where each of the positions L is decomposed into an x-component and a y-component based on a coordinate system having its origin at the ego vehicle 20. Additionally, each respective aggregate difference may be a respective sum of one or more weighted penalties assigned to each of the one or more differences between the respective EM node 36 and the respective characteristic 49 of the prompt node 34.

[0109] At block 190, a check is made as to whether the minimum corresponding difference measure (between the hint node 34 and the best matching node 361) is less than a predetermined matching tolerance. If so, then at block 200, a merge step is performed, wherein the hint node 34 is merged with the best matching EM node 361; but if not (i.e., the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance), then at block 210, a new EM node 361 corresponding to the hint node 34 is merged with the best matching EM node 361. 22 Added to EM structure 38. New EM node 36 22 can be assigned an initial likelihood value (i.e., the initial value of the likelihood LL of the new node); optionally, the new node 36 22 It can also be assigned an initial node risk NR and an initial node reward NW.

[0110] At block 220, the identifying step includes identifying the most likely next node 362 and / or the most dangerous next node 36 10 (Note that although reference numerals 362 and 36 10 exist Figure 3 to denote two specific EM nodes 36, but these reference numerals are also used herein to generally denote the most likely next EM node 36. 22 And the most dangerous next EM node 36 10. ) If the hint node 34 is merged into the best matching node 361, the most likely next node 362 will be the node 36 with the highest likelihood LL and the most dangerous next node 36 among the nodes 36 immediately downstream of the best matching node 361. 10 It will be the node 36 with the highest node risk NR among the nodes 36 immediately downstream of the best matching node 361. On the other hand, if the node 34 is suggested as the new node 36 22 Add, and if immediately following the new node 36 22 If there are one or more nodes 36 downstream, the most likely next node 362 will be immediately after the new node 36. 22 The node 36 with the highest probability LL among the downstream nodes 36 and the most dangerous next node 36 10 will be at the new node 36 22 The node 36 with the highest node risk NR among the downstream nodes 36. 22 There is no node 36 downstream, such as Figure 4 In the case shown in FIG. 3 , the most likely next node 362 or the most dangerous next node 36 will not be identified. 10 However, in this case, autonomous vehicle 20 is experiencing a new event 30 that is significantly different from the previously experienced pre-existing event 40, and as autonomous vehicle 20 proceeds, autonomous vehicle 20 will continue to place new nodes 36 22 38. In this way, EM structure 38 can expand to incorporate all new events 30 experienced by autonomous vehicle 20. EM structure 38 can continue to grow over time and can become very complex such that in some cases, the most likely next node 362 and the most dangerous next node 36 22 The corresponding node risk NR of each EM node 36 may be the corresponding maximum, average, or sum of the corresponding object risks OR of one or more previously observed objects 43 in the associated previously existing event 40. The corresponding object risk OR for each previously observed object 43 may be calculated by applying the corresponding distance d between the previously observed object 43 and the autonomous vehicle 20. 43 The inverse function of 43 exist Figure 5 Used to indicate a specific distance d 43 , but this reference numeral is also used herein to generally represent the distance d between the previously observed object 43 and the autonomous vehicle 20 43 .)For example, Figure 3 and Figure 5 The corresponding distance d to the autonomous vehicle 20 is shown. 43 d44 Since the sigmoidal function returns a value between 0 and 1, where 0 can represent little or no approach risk and 1 can represent high risk, for illustration purposes, assume that the object risk OR for the closer vehicle / object 43 is 43 is 0.7 and is used for the object risk OR of the leading vehicle / object 44 44 is 0.3. In this case, the node risk NR for EM node 36 associated with event 40 may be the two object risks OR 43 , OR 44 The maximum value (0.7), average value (0.5) or sum of these objects. The sum can be the risk OR 43 , OR 44 (1.0), the sum of the maximum and the average (1.2), the midpoint between the maximum and the average (0.6), or any other suitable combination.

[0111] The method 100 may further include the following steps: (i) at box 130, calculating a respective individual risk IR for each of the one or more newly observed objects using a sigmoidal function applied to a respective distance d between each respective newly observed object 21 and the autonomous vehicle 20; (ii) at box 140, establishing a respective overall or total risk for each of the plurality of newly defined regions of interest 32 based on the respective individual risk IR of the one or more newly observed objects 21 located within each respective newly defined region of interest 32; and (iii) at box 150, defining the prompt node 34 as a grouping of the plurality of newly defined regions of interest 32 organized according to their respective overall risk TR. (For example, the grouping of regions of interest 32 may be sorted according to one of the following: low / high / urgent risk regions 32 L , 32 H , 32 U ; Emergency / High / Low Risk Area 32 U , 32 H , 32 L ; and ascending or descending order of overall / total risk TR. )

[0112] The overall / total risk TR for each newly defined region of interest 32 can be determined from the individual risks IR of the newly observed objects 21 in that region of interest 32. For example, using Figure 1 to Figure 2 New event 30 shown in FIG. 32 for emergency risk area U , 32 a5 The overall / total risk TR can be compared with the area 32 U , 32 a5 The individual risk IR of the leading vehicle 22 in the region 32 is the same.U , 32 a5 In the emergency risk area 32 U , 32 a5 The overall / total risk TR will be from 32 in the area of ​​interest U , 32 a5 The individual risks IR of all three vehicles / objects in the region 32 are determined. The overall / total risk TR for the newly defined region of interest 32 may be the maximum, average, sum or aggregation of the individual risks IR of the vehicles / objects in the region 32 .

[0113] The object risk OR for each previously observed object 43 and / or the individual risk IR for each newly observed object 21 can be given by the sigmoidal equation Determine, where R is the corresponding object risk OR or individual risk IR, Mindist is the distance d from the autonomous vehicle 20 to the previously observed object 43 or the newly observed object 21 43 d 21 , and among them SAFEDIST is a distance that depends on the road surface type, road structure type, weather / environmental conditions, and one or more of the relative lane position, closing speed, or closing acceleration between the previously observed object 43 or the newly observed object 21 and the autonomous vehicle 20 .

[0114] The method 100 may further include, at block 180, a step of back-propagating a corresponding node risk NR associated with a high-risk EM node 36 to one or more EM nodes 36 upstream of the high-risk EM node 36, wherein the node risk NR associated with the high-risk EM node 36 is greater than a predetermined risk threshold. This back-propagation of the associated node risk NR may utilize a linear function or a logistic function. For example, if a given EM node 36 has a node risk NR above a predetermined risk threshold, the node may be considered a high-risk node 36, and the node risk NR for the high-risk node 36 may be back-propagated to one or more other nodes 36 immediately upstream of the high-risk node 36.

[0115] Method 100 may further include, at box 110, a step of examining a set of positions L, velocities S, and headings H associated with one or more newly observed objects 21 as sensed data; and, at box 120, a step of converting the sensed data into input, signals, or data associated with a prompt node 34, such as for receipt by a data ingestion module 260 (described in further detail below).

[0116] Figure 8FIG. 2 is a block diagram of a system / controller 230 for predictive navigation control of an autonomous vehicle 20 , which is configured to perform some or all of the steps of the aforementioned method 100 . The system / controller 230 may include Figure 8 All of the blocks or modules shown in the embodiment of the present invention may include, or may include a subset of these blocks or modules. Each of these blocks or modules may be implemented in hardware, software, and / or firmware, wherein two or more of these blocks or modules are integrated together (e.g., as part of a single hardware unit) and one or more other blocks or modules are implemented separately.

[0117] The system / controller 230 includes a data ingestion module 260, an EM module 270, and a prediction module 280. The data ingestion module 260 is configured to receive input associated with a hint node 34, wherein the hint node 34 represents a new event 30 associated with a set of corresponding positions L, speeds S, and headings H associated with one or more newly observed objects 21, wherein each newly observed object 21 is located within a corresponding one of a plurality of newly defined regions of interest 32 surrounding the autonomous vehicle 20. The data ingestion module 260 is further configured to: (i) calculate a corresponding individual risk IR for each of the one or more newly observed objects 21 using a sigmoidal function applied to a corresponding distance d between each corresponding newly observed object 21 and the autonomous vehicle 20; (ii) establish a corresponding overall risk TR for each of the plurality of newly defined regions of interest 32 based on the corresponding individual risk IR of the one or more newly observed objects 21 located within each corresponding newly defined region of interest 32; and (iii) define the hint node 34 as a grouping of the plurality of newly defined regions of interest 32 organized according to their corresponding overall risk TR.

[0118] The EM module 270 is operably connected to the data ingestion module 260 and includes an EM structure 38, which includes a selectively interconnected and directed network of EM nodes 36, wherein each EM node 36 represents a respective pre-existing event 40, wherein each pre-existing event 40 is associated with a respective set of positions L, velocities S, and headings H associated with one or more previously observed objects 43, each previously observed object 43 being located within one of a plurality of previously defined regions of interest 42 surrounding the autonomous vehicle 20, and wherein each EM node 36 has a respective node risk NR and a respective likelihood LL associated therewith. The EM module 270 is configured to: (x) compare the hint node 34 with each of the EM nodes 36; (y) determine which of the EM nodes 36 has a minimum corresponding difference metric, thereby defining a best matching EM node 361, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding EM node 36 and one or more corresponding characteristics 49 of the hint node 34; and (z) merge the hint node 34 with the best matching EM node 361 if the minimum corresponding difference metric is less than a predetermined matching tolerance, otherwise merge a new EM node 361 corresponding to the hint node 34 with the best matching EM node 361 if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance. 22 Added to EM structure 38.

[0119] The prediction module 280 is operatively connected to the EM module 270 and is configured to identify one or both of the following: the next best matching EM node 361 or the new EM node 36 22 the most likely next node 362 among one or more EM nodes 36 downstream, wherein the most likely next node 362 has the highest likelihood LL among the immediately downstream EM nodes 36; and the most likely next node 362 among the immediately downstream EM nodes 36; 22 The most dangerous next node 36 among one or more downstream EM nodes 36 10 , the most dangerous next node 36 10 Among the immediately downstream EM nodes 36 , the node risk NR is the highest.

[0120] The system / controller 230 may further include a perception module 250 that is operably connected to the data ingestion module 260 and is configured to: examine a set of corresponding positions L, speeds S, and headings H associated with one or more newly observed objects 21 as sensed data; and convert the sensed data into input associated with the prompt node 34 for receipt by the data ingestion module 260. In addition, the system / controller 230 may include one or more sensors 240 (e.g., RADAR, LIDAR, ultrasonic, infrared, temperature, etc.) that are operably connected to the perception module 250 and provide input signals to the perception module 250. In addition, the prediction module 280 may be operably connected to other modules / devices 290 external to the system / controller 230.

[0121] The corresponding node risk NR of each EM node 36 may be the corresponding maximum, average, or sum of the corresponding object risks OR of one or more previously observed objects 43 in the associated previously existing event 40, and the corresponding object risk OR for each previously observed object 43 may be calculated by applying the corresponding distance d between the previously observed object 43 and the autonomous vehicle 20. 43 The EM module 270 may be further configured to back-propagate the corresponding node risk NR associated with the high-risk EM node 36 to one or more EM nodes 36 upstream of the high-risk EM node 36, wherein the associated node risk NR is greater than a predetermined risk threshold.

[0122] Although the predictive navigation control for the autonomous vehicle 20 has been described as follows Figures 1 to 8 The method of the present disclosure shown in , however, the method can also be extended to optimize the decision for the decision maker 50. For example, Fig.10 A flow chart of a method 300 for optimizing a decision for a decision maker 50 is shown, wherein the method 300 includes a comparison step (block 360), a determination step (block 370), a merging step (block 400), and an identification step (block 420). The method 300 may also be referred to as Figures 11 to 13 Describe, where Fig.11 A decision maker 50 and various stimuli 51, 52, 53, 54, 55, 56 associated with a new event 30 are shown. Fig.12 The decision maker 50 and the selected stimuli 51, 52, 53 and the selected region of interest 32 are shown. UL , 32 UR , 32 LL , 32 LR ,and Fig.13 39 is used to compare the prompt node 34 with the EM node 36. (Note that the prompt node 34 is shown to be compared with the EM node 36 in the EM structure 38. Figure 4 Also applies to Method 300.)

[0123] The decision maker 50 may be a system or device configured to make decisions, such as a process controller, a control module, etc. The decision maker 50 may use the method 300 to guide, assist, or optimize the decision making process of the decision maker. The stimuli 51-56 may be information related to the new event 30, and the stimuli 51-56 may be visualized or represented as a single point or bit of information in an information space. The information space may be a two-dimensional, three-dimensional, or high-dimensional structure with multiple axes or dimensions. For example, Figure 11 to Figure 12 As shown in , the information space can be visualized or represented as a two-dimensional area with two orthogonal axes that intersect at an origin that can visualize or represent the decision maker 50. Each of these axes can be used to measure a specific characteristic, quality, or feature. For example, the vertical axis can measure the difficulty level and the horizontal axis can measure the cost, with the origin located at a nominal or predetermined combination of the difficulty level and the cost. The sum of the stimuli 51-56 represents the information about the various bits of the event 30 as observed at the current state 31, and the event 30 can be any situation involving or requiring an action or decision. The current state 31 can be the current point in time and can include the state of one or more features of the decision maker 50 at the current point in time. For example, the features of the decision maker 50 can include the contents of the memory, registers, and buffers associated with the decision maker 50, the current command set and target 50 under which the decision maker 50 operates, etc.

[0124] Each stimulus 51 may have one or more attributes or qualities AT associated with it. (Note that although Figures 11 to 13 Reference numeral 51 is used herein to denote a specific newly observed stimulus 51, but this reference numeral is also used herein to denote one or more newly observed stimuli 51 in general. Fig.15 As shown in , these attributes or qualities AT may include importance AT I , Emergency AT U , Risk AT R , Reward AT W , cost AT C , duration AT D and difficulty AT X The attributes or qualities AT may be the same as the characteristics, qualities or features measured or indicated by the axes in the information space, or they may be different therefrom.

[0125] Method 300 involves: taking a given event 30 involving one or more stimuli 51-56 ( Fig.11 ); Identify the arrangement of the region of interest 32 defined relative to the current state 31 of the decision maker 50 ( Fig.12), where the associated stimuli 51-53 are located within and / or assigned to their respective focus regions 32; organizing the collection of focus regions 32 and other data according to event structures 39 ( Fig.13 ); and representing this collection of information as a cue node 34 that can be compared to individual episodic memory nodes 36 in an episodic memory structure 38 ( Fig.14 If the hint node 34 sufficiently matches one of the EM nodes 36 in the EM structure 38, the hint node 34 may be merged or incorporated into the matching EM node 361; however, if the hint node 34 does not sufficiently match any of the EM nodes 36, the hint node 34 may be used as a new EM node 36. 22 Added to EM structure 38.

[0126] Turn again Figures 10 to 13 At block 360, the comparison step includes comparing the hint node 34 with each of the plurality of EM nodes 36 in the EM structure 38. The hint node 34 represents a corresponding set of attributes AT associated with one or more newly observed stimuli 51, 52, 53. 51 , AT 52 , AT 53 associated new events 30, wherein each stimulus 51 is assigned to a plurality of newly defined regions of interest 32 defined relative to the current state 31 of the decision maker 50 L , 32 H , 32 U The EM structure 38 comprises a selectively interconnected and directed network of EM nodes 36, wherein each EM node 36 represents a corresponding pre-existing event 40. Each pre-existing event 40 is associated with a corresponding set of attributes AT associated with one or more previously observed stimuli 61, 62, 63. 61 , AT 62 , AT 63 In association, each stimulus is assigned to a plurality of previously defined regions of interest 42 defined relative to a corresponding previous state 41 of the decision maker 50 L , 42 H , 42 U . (The previous state 41 may be a previous moment in time, and may include the state of one or more features of the decision maker 50 at the previous moment in time.) Each EM node 36 has a corresponding node risk NR and a corresponding likelihood LL associated therewith, and may optionally have a corresponding node reward NW associated therewith. (Note that although Figures 11 to 13 Reference numeral 61 is used in the text to denote a specific previously observed stimulus 61, but this reference numeral is also used herein to denote one or more previously observed stimuli 61 generally.)

[0127] The goals G of the decision maker 50 may include one or more of minimizing risk GMR, optimizing utility GOU, optimizing risk and reward GRR, learning to anticipate the needs of the user GAN, finding the most useful information GFI, and learning how to recognize when to provide help to the user GOH. These goals G may be useful when the decision maker 50 is used in a process controller, a logic / control module, a personal data assistant, a virtual assistant, an expert system, etc. Each of the corresponding attributes AT of one or more newly observed stimuli 51 and one or more previously observed stimuli 61 may be defined relative to one or more goals G of the decision maker 50. For example, the goals G may determine the axes used in the information space, and the stimuli 51, 61 may be placed, visualized, and / or represented relative to these axes in the information space.

[0128] Focus area 32 Fig.12 32 as a quadrant representation of an elliptical subset of the information space. Note that other possible shapes and sizes of subsets of the information space may also be used. The stimulus 51 is in the lower left quadrant or the region of interest 32 LL As shown in FIG. 3 , the region of interest 32 LL Shown as emergency risk area 32 U (Since stimulus 51 is close to the origin). Stimulus 52 is in the upper left quadrant or area of ​​interest 32 UL As shown in FIG. 3 , the region of interest 32 UL Shown as high risk area 32 H , and stimulus 53 is in the lower right quadrant or area of ​​interest 32 LR As shown in FIG. 3 , the region of interest 32 LR Also shown as high risk area 32 H . Upper right quadrant or area of ​​interest 32 UR There is no stimulus and it is shown as a low risk area in light grey shading 32 L Note that stimuli 54, 55 and 56 are not shown because they do not fall within any region of interest 32. Stimuli 51, 52 and 53 have corresponding influence potentials ep 51 ep 52 and ep 53 , which represents the effect that each stimulus 51, 52, 53 may have on one or more goals G of the decision maker 50, as further described below. The distance between the origin and a given stimulus 51 represents the magnitude of the effect of the stimulus on achieving one or more goals G. For example, since stimulus 51 is closer to the origin than stimulus 53, stimulus 51 may be considered to have a greater effect on achieving (or not achieving) goal G than stimulus 53.

[0129] At block 370, the determining step includes determining which of the plurality of EM nodes 36 has the smallest corresponding difference metric, thereby defining the best matching EM node 361. Each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding EM node 36 and one or more corresponding aspects AS of the prompt node 34. These aspects AS are similar to feature 50 of method 100 and are described in detail below. Fig.15 As shown in , it can include user temperament AST (e.g., mood, energy level, alertness level, etc.), user skill level AS U (e.g., based on previous decisions and outcomes vs. goal G), environmental factors AS E (e.g., time of day or night, level of distractions in the surrounding environment, etc.) and previous user trends and habits AS H (based on previous representations) and the corresponding importance AS associated with one or more newly observed stimuli 51 or previously observed stimuli 61 I 、UrgencyAS U 、Risk AS R 、Reward AS W , cost AS C , duration AS D and difficulty AS X Note that some aspects AS can also be attributes AT, and vice versa, such as Fig.15 The importance can be expressed as the aspect AS I and / or attribute AT I . Emergency AS U , AT U 、Risk AS R , AT R 、Reward AS W , AT W , cost AS C , AT C , duration AS D , AT D and difficulty AS X , AT X The same is true for . In some cases, these corresponding aspects AS and attributes AT can be used interchangeably. Note that although Fig.13 Only one attribute AT is shown for each newly observed stimulus 51, 52, 53 (i.e., AT 51 , AT 52 , AT 53 ) and one attribute AT for each previously observed stimulus 61, 62, 63 (i.e., AT 61 , AT 62 , AT 63 ) and an aspect AS for the new event 30 (ie AS30 ) and an aspect AS for a pre-existing event 40 (ie AS 40 ), but there can be more than one attribute AT or aspect AS associated with each of them.

[0130] The aggregate difference may be a sum, average, or maximum of the differences between or among the respective aspects of the hint node 34 and each EM node 36. Additionally, each respective aggregate difference may be a respective sum of one or more weighted penalties assigned to each of the one or more differences between the respective EM node 36 and the respective aspect of the hint node 34.

[0131] At block 390, a check is made as to whether the minimum corresponding difference measure (between the hint node 34 and the best matching node 361) is less than a predetermined matching tolerance. If so, then at block 400, a merge step is performed, wherein the hint node 34 is merged with the best matching EM node 361; but if not (i.e., the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance), then at block 410, a new EM node 361 corresponding to the hint node 34 is merged with the best matching EM node 361. 22 Added to EM structure 38. New EM node 36 22 can be assigned an initial likelihood value (i.e., the initial value of the likelihood LL of the new node); optionally, the new node 36 22 It can also be assigned an initial node risk NR and an initial node reward NW.

[0132] At block 420, the identifying step includes identifying the most likely next node 362 and / or the most dangerous next node 36 10 If the hint node 34 is merged into the best matching node 361, the most likely next node 362 will be the node 36 with the highest likelihood LL among the nodes 36 immediately downstream of the best matching node 361. On the other hand, if the hint node 34 is used as the new node 36 22 Add, and if immediately following the new node 36 22 If there are one or more nodes 36 downstream of the new node, the most likely next node 362 will be the node immediately following the new node 36. 22 The node 36 with the highest probability LL among the downstream nodes 36 and the most dangerous next node 36 10 will be at the new node 36 22 The node 36 with the highest node risk NR among the nodes 36 downstream of Figure 4 In the case shown in FIG. 36, immediately following the new node 36 22 If there is no node 36 downstream of 10However, in this case, it is possible that the decision maker 50 is experiencing a new event 30 that is significantly different from the previously experienced pre-existing event 40, and as the decision maker 50 progresses, the decision maker 50 will continue to place the new node 36 22 38. In this way, the EM structure 38 can expand to incorporate all new events 30 experienced by the decision maker 50. The EM structure 38 can continue to grow over time and can become very complex so that in some cases the most likely next node 362 and the most dangerous next node 36 22 It may be the same node 36.

[0133] The corresponding node risk NR of each EM node 36 may be the corresponding stimulus risk SR of one or more previously observed stimuli 61, 62, 63 in the associated pre-existing event 40. 61 , SR 62 , SR 63 The corresponding maximum value, average value or sum of the corresponding stimulus risk SR for each previously observed stimulus 61, 62, 63 can be obtained by applying the corresponding influence potential ep between the previously observed stimulus 61, 62, 63 and one or more targets G of the decision maker 50. 61 ,ep 62 ,ep 63 The inverse function of is determined. (Note that although the reference numeral ep 61 exist Fig.12 Used in EP to indicate specific impact potential 61 , but this reference numeral is also used herein to generally represent the influence potential ep between a previously observed stimulus 61 and one or more targets G of the decision maker 50 61 . )

[0134] The method 300 may further include, at block 330, the following steps: (i) using the corresponding influence potential ep between each corresponding newly observed stimulus 51, 52, 53 and one or more targets G of the decision maker 50 51 ,ep 52 ,ep 53 The sigmoidal function of IR is calculated for each of the one or more newly observed stimuli 51, 52, 53. 51 IR 52 IR 53; (ii) at box 340, establishing a corresponding overall risk TR for each of the multiple newly defined areas of interest 32 based on the corresponding individual risk IR of one or more newly observed stimuli 51 assigned to each corresponding newly defined area of ​​interest 32; and (iii) at box 350, defining the prompt node 34 as a grouping of the multiple newly defined areas of interest 32 organized according to their corresponding overall risk TR.

[0135] The method 300 may further include, at block 380, a step of back-propagating a corresponding node risk NR associated with a high-risk EM node 36 to one or more EM nodes 36 upstream of the high-risk EM node 36, wherein the node risk NR associated with the high-risk EM node 36 is greater than a predetermined risk threshold. This back-propagation of the associated node risk NR may utilize a linear function or a logistic function. For example, if a given EM node 36 has a node risk NR above a predetermined risk threshold, the node may be considered a high-risk node 36, and the node risk NR for the high-risk node 36 may be back-propagated to one or more other nodes 36 immediately upstream of the high-risk node 36.

[0136] Method 300 may further include, at box 310, a step of examining, as perceived data, a set of corresponding attributes AT associated with one or more newly observed stimuli 51; and, at box 320, a step of converting the perceived data into input, signal or data associated with a prompt node 34, such as for receipt by a data ingestion module 460 (described in further detail below).

[0137] Fig. 9 A block diagram of a system / controller 430 for optimizing decisions for the decision maker 50 is shown. The system / controller 430 is configured to perform some or all of the steps of the aforementioned method 300. The system / controller 430 may include Fig. 9 All of the blocks or modules shown in the embodiment of the present invention may include, or may include a subset of these blocks or modules. Each of these blocks or modules may be implemented in hardware, software, and / or firmware, wherein two or more of these blocks or modules are integrated together (e.g., as part of a single hardware unit) and one or more other blocks or modules are implemented separately.

[0138] The system / controller 430 includes a data ingestion module 460, an EM module 470, and a prediction module 480. The data ingestion module 460 is configured to receive input associated with a prompt node 34, wherein the prompt node 34 represents a new event 30 associated with a set of corresponding attributes AT associated with one or more newly observed stimuli 51, wherein each newly observed stimulus 51 is located within and / or assigned to a corresponding one of a plurality of newly defined areas of interest 32 defined relative to a current state 31 of the decision maker 51. The data ingestion module 460 is further configured to: (i) calculate a corresponding individual risk IR for each of the one or more newly observed stimuli 51 using a sigmoidal function of a corresponding influence potential ep between each corresponding newly observed stimulus 51 and one or more targets G of the decision maker 50; (ii) establish a corresponding overall risk TR for each of the multiple newly defined regions of interest 32 based on the corresponding individual risk IR of the one or more newly observed stimuli 51 located within each corresponding newly defined region of interest 32 and / or assigned to each corresponding newly defined region of interest 32; and (iii) define the prompt node 34 as a grouping of the multiple newly defined regions of interest 32 organized according to their corresponding overall risk TR.

[0139] The EM module 470 is operably connected to the data ingestion module 460 and includes an EM structure 38, which includes a selectively interconnected and directed network of EM nodes 36, wherein each EM node 36 represents a respective pre-existing event 40, wherein each pre-existing event 40 is associated with a respective set of attributes AT associated with one or more previously observed stimuli 61, each previously observed stimulus 61 is located within and / or assigned to a respective one of a plurality of previously defined regions of interest 42, and wherein each EM node 36 has a respective node risk NR and a respective likelihood LL associated therewith. The EM module 470 is configured to: (x) compare the hint node 34 with each of the EM nodes 36; (y) determine which of the EM nodes 36 has a minimum corresponding difference metric, thereby defining a best matching EM node 361, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding EM node 36 and one or more corresponding aspects AS of the hint node 34; and (z) merge the hint node 34 with the best matching EM node 361 if the minimum corresponding difference metric is less than a predetermined matching tolerance, otherwise merge a new EM node 361 corresponding to the hint node 34 with the best matching EM node 361 if the minimum corresponding difference metric is greater than or equal to the predetermined matching tolerance. 22 Added to EM structure 38.

[0140] The prediction module 480 is operatively connected to the EM module 470 and is configured to identify one or both of the following: the next best matching EM node 361 or the new EM node 36 22 the most likely next node 362 among one or more EM nodes 36 downstream, wherein the most likely next node 362 has the highest likelihood LL among the immediately downstream EM nodes 36; and the most likely next node 362 among the immediately downstream EM nodes 36; 22 The most dangerous next node 36 among one or more downstream EM nodes 36 10 , the most dangerous next node 36 10 The node with the highest node risk NR is among the immediately downstream EM nodes 36. In addition, the prediction module 480 may be operably connected with other modules / devices 490 external to the system / controller 430.

[0141] The system / controller 430 may further include a perception module 450 that is operably connected to the data ingestion module 460 and is configured to: examine, as sensed data, a set of corresponding attributes AT associated with one or more newly observed stimuli 51; and convert the sensed data into input associated with the prompt node 34 for receipt by the data ingestion module 460. In addition, the system / controller 430 may include one or more sensors 440 (e.g., RADAR, LIDAR, ultrasonic, infrared, temperature, etc.) that are operably connected to the perception module 450 and provide input signals to the perception module 450.

[0142] The corresponding node risk NR of each EM node 36 may be a corresponding maximum, average, or sum of corresponding object risks OR of one or more previously observed stimuli 61 in the associated previously existing events 40, and the corresponding object risk OR for each previously observed stimulus 61 may be determined by a sigmoidal function applied to a corresponding impact potential ep between the previously observed object 43 and one or more targets G of the decision maker 50. The EM module 470 may be further configured to back-propagate the corresponding node risk NR associated with the high-risk EM node 36 to one or more EM nodes 36 upstream of the high-risk EM node 36, where the associated node risk NR is greater than a predetermined risk threshold.

[0143] Ability to predict the next most dangerous EM node 36 10 The most likely next EM node 362 may assist the autonomous vehicle 20 in making its next action or decision (thereby providing a form of predictive navigation), and may likewise assist the decision maker 50 in optimizing its next action or decision.

[0144] The above description is intended to be illustrative, not restrictive. Although the size and type of the material described herein are intended to be illustrative, they are by no means restrictive and are exemplary embodiments. In the following claims, the use of the terms "first", "second", "top", "bottom", etc. is used only as a label and is not intended to impose numerical or positional requirements on its object. As used herein, elements or steps stated in singular form and added before the word "one" or "one" should be understood as not excluding the plural number of such elements or steps, unless such exclusion is explicitly indicated. In addition, the phrase "at least one of A and B" and the phrase "A and / or B" should each be understood to refer to "only A, only B, or both A and B". In addition, unless there is a clear statement to the contrary, embodiments that "include" or "have" one or more elements with a specific characteristic may include additional such elements that do not have the characteristic. And when broad descriptive adverbs such as "substantially" and "generally" are used to modify adjectives in this document, these adverbs mean "mostly", "to a significant extent" and / or "to a large extent", and do not necessarily mean "completely", "thoroughly", "strictly" or "entirely".

[0145] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functions and / or operations of possible implementations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flowchart or block diagram may represent a module, fragment or portion of a code, which includes one or more executable instructions for implementing a specified logical function. It will be noted that each box illustrated in the block diagram and / or flowchart and the combination of boxes in the block diagram and / or flowchart can be implemented by a hardware-based system that performs a specified function or action, or a combination of hardware and computer instructions. These computer program instructions can also be stored in a computer-readable medium, which can instruct a controller or other programmable data processing device to act in a specific manner so that the instructions stored in the computer-readable medium produce an article of manufacture, which includes instructions for implementing the functions and / or actions specified in the flowchart and block diagram.

[0146] This written description uses examples, including the best mode, to enable those skilled in the art to make and use devices, systems, and components of objects, and to perform methods, in accordance with the present disclosure. It is the following claims, including equivalents, that define the scope of the present disclosure.

Claims

1. A method for predictive navigation control of an autonomous vehicle, the method comprising: comparing the cue node to each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event, the new event being associated with a corresponding set of positions, velocities, and headings associated with one or more newly observed objects, wherein each object is located within a corresponding one of a plurality of newly defined zones of interest surrounding the autonomous vehicle, and wherein the episodic memory structure comprises a selectively interconnected and directed network of the episodic memory nodes, wherein each episodic memory node represents a corresponding pre-existing event, wherein each pre-existing event is associated with a corresponding set of positions, velocities, and headings associated with one or more previously observed objects, wherein each object is located within a corresponding one of a plurality of previously defined zones of interest surrounding the autonomous vehicle, and wherein each episodic memory node has a corresponding node risk and a corresponding likelihood associated therewith; determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between one or more corresponding characteristics of the corresponding episodic memory node and the cue node; if the minimum corresponding difference measure is less than a predetermined matching tolerance, merging the prompt node with the best matching episodic memory node, or if the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance, adding a new episodic memory node corresponding to the prompt node to the episodic memory structure; as well as Identifying one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most risky next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

2. The method of claim 1 , wherein the corresponding node risk of each contextual memory node is a corresponding maximum, average, or sum of corresponding object risks of the one or more previously observed objects in an associated previously existing event, and wherein the corresponding object risk for each previously observed object is determined by a sigmoidal function applied to the corresponding distance between the previously observed object and the autonomous vehicle.

3. The method according to claim 2, further comprising: calculating a respective individual risk for each of the one or more newly observed objects using the sigmoidal function applied to a respective distance between each respective newly observed object and the autonomous vehicle; establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed objects located within each respective newly defined region of interest; as well as The prompt nodes are defined as groupings of the plurality of newly defined areas of interest organized according to their respective overall risks.

4. The method of claim 3 , wherein the corresponding object risk or individual risk for each previously observed or newly observed object, respectively, is determined by R=2·({1-1 / (1+e^[(mindist-SAFEDIST) / (SAFEDIST / 2)]}-0.5)+0.5, wherein R is the corresponding object risk or individual risk, mindist is the distance from the autonomous vehicle to the previously observed or newly observed object, and wherein SAFEDIST is a distance that depends on one or more of a road surface type, a road structure type, weather / environmental conditions, and a relative lane position, a closing speed, or a closing acceleration between the previously observed or newly observed object and the autonomous vehicle.

5. The method of claim 1, wherein each of the respective positions, velocities, and headings of the one or more newly observed objects and the one or more previously observed objects are defined relative to the autonomous vehicle.

6. The method of claim 1 , wherein each respective aggregate difference is a respective sum of one or more weighted penalties assigned to each of one or more differences between the respective characteristics of the respective contextual memory node and the cue node, and wherein the characteristics include one or more of: number and type of regions of interest, number of objects in each region of interest, road surface type, road structure type, environment type, weather / environmental conditions, driving goal, driving mode, vehicle type, powertrain type, and respective position, speed, heading, distance from the autonomous vehicle, and object risk associated with the one or more previously observed and / or newly observed objects.

7. The method according to claim 1, further comprising: Back-propagating a corresponding node risk associated with a high-risk episodic memory node to one or more episodic memory nodes upstream of the high-risk episodic memory node, wherein the associated node risk is greater than a predetermined risk threshold.

8. The method of claim 7, wherein the back-propagation of the associated node risks utilizes a linear function or a logistic function.

9. The method of claim 1 , wherein when a new episodic memory node is added to the episodic memory structure, the new episodic memory node is added as a child node to one or more parent nodes, wherein each parent node is a previously existing episodic memory node, and wherein the new episodic memory node is assigned an initial likelihood value.

10. The method according to claim 1, wherein each contextual memory node also has a corresponding node reward.

11. A method for optimizing decision making for a decision maker, the method comprising: comparing the cue node to each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event associated with a corresponding set of attributes associated with one or more newly observed stimuli, each stimulus assigned to a corresponding one of a plurality of newly defined attention regions defined relative to a current state of the decision maker, and wherein the episodic memory structure comprises a selectively interconnected and directed network of the episodic memory nodes, wherein each episodic memory node represents a corresponding pre-existing event, wherein each pre-existing event is associated with a corresponding set of attributes associated with one or more previously observed stimuli, wherein each stimulus assigned to a corresponding one of a plurality of previously defined attention regions defined relative to a corresponding previous state of the decision maker, and wherein each episodic memory node has a corresponding node risk and a corresponding likelihood associated therewith; determining which of the plurality of episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between the corresponding episodic memory node and one or more corresponding aspects of the cue node; if the minimum corresponding difference measure is less than a predetermined matching tolerance, merging the prompt node with the best matching episodic memory node, or if the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance, adding a new episodic memory node corresponding to the prompt node to the episodic memory structure; as well as Identifying one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most risky next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

12. The method of claim 11, wherein each of the corresponding attributes of the one or more newly observed stimuli and the one or more previously observed stimuli are defined relative to one or more objectives of the decision maker.

13. A method according to claim 12, wherein the corresponding node risk of each episodic memory node is the corresponding maximum, average or sum of the corresponding stimulus risks of the one or more previously observed stimuli in the associated pre-existing events, and wherein the corresponding stimulus risk for each previously observed stimulus is determined by a sigmoidal function applied to the corresponding influence potential between the previously observed stimuli and the one or more targets of the decision maker.

14. The method according to claim 13, further comprising: calculating a respective individual risk for each of the one or more newly observed stimuli using the sigmoidal function applied to the respective impact potential between each respective newly observed stimulus and the one or more objectives of the decision maker; establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed stimuli assigned to each respective newly defined region of interest; as well as The prompt nodes are defined as groupings of the plurality of newly defined areas of interest organized according to their respective overall risks.

15. The method of claim 11, wherein each respective aggregate difference is a respective sum of one or more weighted penalties assigned to each of the one or more differences between the respective aspect of the respective episodic memory node and the cue node.

16. The method of claim 12, wherein: (i) the attributes include one or more of importance, urgency, risk, reward, cost, duration and difficulty; (ii) the aspects include user temperament, user skill level, environmental factors, previous user habits and one or more of the corresponding importance, urgency, risk, reward, cost, duration and difficulty associated with the one or more previously observed or newly observed stimuli; and (iii) the goals of the decision maker include one or more of minimizing risk, optimizing utility, optimizing risk and reward, learning to anticipate user needs, finding the most useful information and learning how to recognize when to provide help to the user.

17. A controller for predictive navigation control of an autonomous vehicle, the controller comprising: (a) a data ingestion module configured to receive input associated with a hint node, wherein the hint node represents a new event associated with a set of respective positions, velocities, and headings associated with one or more newly observed objects, each of the newly observed objects being located within a respective one of a plurality of newly defined regions of interest surrounding the autonomous vehicle, the data ingestion module being further configured to: calculating a respective individual risk for each of the one or more newly observed objects using a sigmoidal function applied to a respective distance between each respective newly observed object and the autonomous vehicle; establishing a respective overall risk for each of the plurality of newly defined regions of interest based on the respective individual risks of the one or more newly observed objects located within each respective newly defined region of interest; as well as defining the alert nodes as a grouping of the plurality of newly defined areas of interest organized according to their respective overall risks; (b) an episodic memory module operably connected to the data ingestion module and comprising an episodic memory structure, the episodic memory structure comprising a selectively interconnected and directed network of episodic memory nodes, wherein each episodic memory node represents a respective pre-existing event, wherein each pre-existing event is associated with a respective set of positions, velocities, and headings associated with one or more previously observed objects, wherein each previously observed object is located within a respective one of a plurality of previously defined regions of interest surrounding the autonomous vehicle, and wherein each episodic memory node has a respective node risk and a respective likelihood associated therewith, the episodic memory module being configured to: comparing the prompt node with each of the episodic memory nodes; determining which of the episodic memory nodes has a minimum corresponding difference metric, thereby defining a best matching episodic memory node, wherein each corresponding difference metric is determined based on a corresponding aggregate difference between one or more corresponding characteristics of the corresponding episodic memory node and the cue node; as well as merging the prompt node with the best matching episodic memory node if the minimum difference measure is less than a predetermined matching tolerance, or adding a new episodic memory node corresponding to the prompt node to the episodic memory structure if the minimum corresponding difference measure is greater than or equal to the predetermined matching tolerance; and (c) a prediction module operably connected to the episodic memory module and configured to identify one or both of: (i) a most likely next episodic memory node among one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most likely next episodic memory node has the highest likelihood among the immediately downstream episodic memory nodes, and (ii) a most risky next episodic memory node among the one or more episodic memory nodes immediately downstream of the best match or new episodic memory node, wherein the most risky next episodic memory node has the highest node risk among the immediately downstream episodic memory nodes.

18. The controller according to claim 17, further comprising: (d) a perception module, the perception module being operatively connected to the data ingestion module and configured to: detecting, as sensed data, the set of corresponding positions, velocities, and headings associated with the one or more newly observed objects; and The sensed data is converted into the input associated with the prompt node for receipt by the data ingestion module.

19. The controller of claim 17 , wherein the corresponding node risk for each contextual memory node is a corresponding maximum, average, or sum of corresponding object risks for the one or more previously observed objects in the associated pre-existing events, and wherein the corresponding object risk for each previously observed object is determined by the sigmoidal function applied to the corresponding distance between the previously observed object and the autonomous vehicle.

20. The controller of claim 17, wherein the context memory module is further configured: Back-propagating a corresponding node risk associated with a high-risk episodic memory node to one or more episodic memory nodes upstream of the high-risk episodic memory node, wherein the associated node risk is greater than a predetermined risk threshold.

Images