System and method for predictive navigation control

Patent metadata

Field	Value
Publication number	US-11485387-B2
Application number	US-202017114977-A
Country	US
Kind code	B2
Filing date	Dec 8, 2020
Priority date	Dec 8, 2020
Publication date	Nov 1, 2022
Grant date	Nov 1, 2022

Abstract

A method of predictive navigation control for an ego vehicle includes: comparing a cue node to each of a plurality of episodic memory nodes in an episodic memory structure, wherein the cue node represents a new event representing distances, speeds and headings of one or more newly observed objects about the ego vehicle, and wherein the episodic memory structure includes a network of nodes each representing a respective previously existing event and having a respective node risk and likelihood; determining which of the nodes has a smallest respective difference metric, thus defining a best matching node; consolidating the cue node with the best matching node if the smallest difference metric is less than a match tolerance, else adding a new node corresponding to the cue node to the episodic memory structure; and identifying a likeliest next node and/or a riskiest next node.

First claim

What is claimed is:

1. A method of predictive navigation control for an ego vehicle, comprising:
comparing a 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 set of respective locations, speeds and headings associated with one or more newly observed objects each located within a respective one of a plurality of newly defined attention zones about the ego 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 respective previously existing event, with each previously existing event being associated with a respective set of locations, speeds and headings associated with one or more previously observed objects each located within a respective one of a plurality of previously defined attention zones about the ego vehicle, and wherein each episodic memory node has a respective node risk and a respective likelihood associated therewith;
determining which of the plurality of episodic memory nodes has a smallest respective difference metric, thus defining a best matching episodic memory node, wherein each respective difference metric is determined based on a respective aggregate difference between one or more respective characteristics of the respective episodic memory node and the cue node;
consolidating the cue node with the best matching episodic memory node if the smallest respective difference metric is less than a predetermined match tolerance, or adding a new episodic memory node corresponding to the cue node to the episodic memory structure if the smallest respective difference metric is greater than or equal to the predetermined match tolerance; and
identifying one or both of (i) a likeliest next episodic memory node among one or more episodic memory nodes immediately downstream from the best matching or new episodic memory node, wherein the likeliest next episodic memory node has a highest likelihood among the immediately downstream episodic memory nodes, and (ii) a riskiest next episodic memory node among the one or more episodic memory nodes immediately downstream from the best matching or new episodic memory node, wherein the riskiest next episodic memory node has a highest node risk among the immediately downstream episodic memory nodes.

2. The method according to claim 1 , wherein the respective node risk of each episodic memory node is a respective maximum, average or aggregate of respective object risks for the one or more previously observed objects in the associated previously existing event, and wherein the respective object risk for each previously observed object is determined by a sigmoidal function applied to a respective distance between the previously observed object and the ego 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 ego vehicle;
establishing a respective overall risk for each of the plurality of newly defined attention zones, based on the respective individual risks of the one or more newly observed objects located within each respective newly defined attention zone; and
defining the cue node as a grouping of the plurality of newly defined attention zones organized according to their respective overall risks.

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

5. The method according to claim 1 , wherein each of the respective locations, speeds and headings of the one or more newly observed objects and the one or more previously observed objects is defined with respect to the ego vehicle.

6. The method according to claim 1 , wherein each respective aggregate difference is a respective total of one or more weighted penalties assigned against each of one or more differences between the respective characteristics of the respective episodic memory node and the cue node, and wherein the characteristics include one or more of number and type of attention zones, number of objects in each attention zone, road surface type, road structure type, environment type, weather/environmental conditions, driving goal, driving mode, vehicle type, powertrain type and respective locations, speeds, headings, distances from the ego vehicle and object risks 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 respective 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 according to claim 7 , wherein the back-propagation of the associated node risk utilizes a linear function or a logistic function.

9. The method according to 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 episodic memory node also has a respective node reward.

11. A method of optimizing a decision for a decider, comprising:
comparing a 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 set of respective attributes associated with one or more newly observed stimuli each assigned to a respective one of a plurality of newly defined attention zones defined with respect to a current state of the decider, 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 respective previously existing event, with each previously existing event being associated with a respective set of attributes associated with one or more previously observed stimuli each assigned to a respective one of a plurality of previously defined attention zones defined with respect to a respective previous state of the decider, and wherein each episodic memory node has a respective node risk and a respective likelihood associated therewith;
determining which of the plurality of episodic memory nodes has a smallest respective difference metric, thus defining a best matching episodic memory node, wherein each respective difference metric is determined based on a respective aggregate difference between one or more respective aspects of the respective episodic memory node and the cue node;
consolidating the cue node with the best matching episodic memory node if the smallest respective difference metric is less than a predetermined match tolerance, or adding

Assignees

• Gm Global Tech Operations Llc

Inventors

• Howard Michael D
• Kwon Hyukseong
• Bhattacharyya Rajan

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/045

  Combinations of networks · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• B60W40/09

  Driving style or behaviour · CPC title

• B60W60/0027 Primary

  using trajectory prediction for other traffic participants · CPC title