Introspective autonomous vehicle operational management

• US 10,649,453 B1
• Filed: 11/15/2018
• Issued: 05/12/2020
• Est. Priority Date: 11/15/2018
• Status: Active Grant

• Current Assignee: Nissan Motor Co., Ltd., Renault Sas (Renault), University Of Massachusetts
• Original Assignee: Nissan North America Incorporated (Nissan Motor Co., Ltd.)
• Inventors: Svegliato, Justin, Witwicki, Stefan, Wray, Kyle Hollins, Zilberstein, Shlomo
• Primary Examiner(s): Lee, Tyler J
• Application Number:
  US16/192,356
• Publication Number:
  US 20200159213A1
• Time in Patent Office: 544 Days
• CPC Class Codes: B60W 2050/0297 Control Giving priority to different actuators or systems
• CPC Class Codes: B60W 50/029 Adapting to failures or work around with other constraints, e.g. circumvention by avoiding use of failed parts
• CPC Class Codes: B60W 60/00 Drive control systems specially adapted for autonomous road vehicles
• CPC Class Codes: G05D 1/0077 using redundant signals or controls
• CPC Class Codes: G05D 1/0088 characterized by the autonomous decision making process, e.g. artificial intelligence, predefined behaviours using knowledge based models G06N5/00

Abstract

Introspective autonomous vehicle operational management includes operating an introspective autonomous vehicle operational management controller including a policy for a model of an introspective autonomous vehicle operational management domain. Operating the controller includes, in response to a determination that a current belief state of the policy indicates an exceptional condition, identifying an exception handler for controlling the autonomous vehicle. Operating the controller includes, in response to a determination that the current belief state indicates an unexceptional condition, identifying a primary handler as the active handler. Operating the controller includes controlling the autonomous vehicle to traverse a current portion of the vehicle transportation network in accordance with the active handler, receiving an indicator output by the active handler, generating an updated belief state based on the indicator, and controlling the autonomous vehicle to traverse a subsequent portion of the vehicle transportation network based on the updated belief state.

Claims

1. A method for use in traversing a vehicle transportation network, the method comprising:

traversing, by an autonomous vehicle, a vehicle transportation network, wherein traversing the vehicle transportation network includes;

operating an introspective autonomous vehicle operational management controller, wherein the introspective autonomous vehicle operational management controller includes an introspective autonomous vehicle operational management policy for an introspective autonomous vehicle operational management model of an introspective autonomous vehicle operational management domain, wherein the introspective autonomous vehicle operational management model includes;

a defined set of exceptions; and

a defined set of available handlers; and

wherein operating the introspective autonomous vehicle operational management controller includes the introspective autonomous vehicle operational management controller;

determining whether a current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an exceptional condition from the set of exceptions;

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates the exceptional condition;

identifying an exception handler from the set of available handlers as an active handler for controlling the autonomous vehicle, wherein identifying the exception handler includes;

 

in response to a determination that the set of available handlers includes an internal exception handler that models the exceptional condition, identifying the internal exception handler as the active handler; and

 

in response to a determination that the set of available handlers omits the internal exception handler that models the exceptional condition, identifying an external assistance exception handler as the active handler;

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an unexceptional condition, identifying a primary handler from the set of available handlers as the active handler for controlling the autonomous vehicle;

controlling the autonomous vehicle to traverse a current portion of the vehicle transportation network in accordance with the active handler; and

in response to receiving an indicator from the active handler subsequent to controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the active handler;

generating an updated exception belief state probability distribution based on the current exception belief state probability distribution and the indicator; and

traversing a subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution.

2. The method of claim 1, wherein operating the introspective autonomous vehicle operational management controller includes the introspective autonomous vehicle operational management controller:

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an uncertainty value that exceeds a defined uncertainty threshold;

identifying a vehicle control action for traversing the current portion of the vehicle transportation network;

controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the vehicle control action;

in response to controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the vehicle control action, generating an updated exception belief state probability distribution that indicates an uncertainty value that is within the defined uncertainty threshold; and

traversing a subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution.

3. The method of claim 1, wherein the primary handler includes a model of a defined limited autonomous vehicle operational management domain.

4. The method of claim 3, wherein the defined limited autonomous vehicle operational management domain is a route planning domain and the model from the primary handler is a route planning model of the route planning domain.

5. The method of claim 4, wherein the primary handler includes a scenario-specific operational control evaluation model of a distinct vehicle operational scenario.

6. The method of claim 3, wherein:

the set of exceptions includes one or more exceptional factors that describe exceptional conditions that probabilistically affect the operation of the autonomous vehicle, wherein the model of the defined limited autonomous vehicle operational management domain omits a description of the exceptional conditions.

7. The method of claim 1, wherein the introspective autonomous vehicle operational management model includes a defined set of indicators, wherein the defined set of indicators includes the indicator, and wherein:

identifying the primary handler from the set of available handlers as the active handler includes identifying the primary handler from the set of available handlers as the active handler such that the primary handler outputs the indicator from the defined set of indicators in accordance with a defined temporal term;

identifying the exception handler from the set of available handlers as the active handler includes identifying the exception handler from the set of available handlers as the active handler such that the exception handler outputs the indicator from the defined set of indicators in response to identifying a termination condition, wherein the indicator is;

a success indicator indicating that the exceptional condition is resolved;

or

on a condition that the exception handler is the internal exception handler, a failure indicator indicating that a set of available actions available to the exception handler omits an available action for resolving the exceptional condition.

8. The method of claim 7, wherein the introspective autonomous vehicle operational management model includes:

a defined set of introspective states based on a defined set of states and the defined set of exceptions;

a defined set of introspective actions based on a defined set of actions and the defined set of available handlers;

a transition profile representing a probability of a state from the defined set of states subsequent to traversing the vehicle transportation network from a current exceptional state from the defined set of introspective states in accordance with the active handler;

an exception profile representing a probability of an exception from the set of exceptions subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

a cost profile representing an expected immediate cost subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

an indicator profile representing a probability of observing an indicator from the defined set of indicators subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

an introspective transition function based on the transition profile and the exception profile;

an introspective reward function based on the cost profile and the defined set of available handlers;

a set of introspective observations based on a defined set of observations and the defined set of indicators;

an introspective observation function based on the indicator profile and the defined set of available handlers; and

a defined set of target states.

9. The method of claim 1, wherein, traversing the subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution includes:

on a condition that the active handler is the primary handler and in response to a determination that the updated exception belief state probability distribution indicates an exceptional condition;

suspending the primary handler; and

identifying an exception handler from the set of available handlers as the active handler.

10. An autonomous vehicle comprising:

a processor configured to execute instructions stored on a non-transitory computer readable medium to control the autonomous vehicle to traverse a vehicle transportation network by;

operating an introspective autonomous vehicle operational management controller, wherein the introspective autonomous vehicle operational management controller includes an introspective autonomous vehicle operational management policy for an introspective autonomous vehicle operational management model of an introspective autonomous vehicle operational management domain, wherein the introspective autonomous vehicle operational management model includes;

a defined set of exceptions; and

a defined set of available handlers; and

wherein operating the introspective autonomous vehicle operational management controller includes the introspective autonomous vehicle operational management controller;

determining whether a current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an exceptional condition from the set of exceptions;

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates the exceptional condition;

identifying an exception handler from the set of available handlers as an active handler for controlling the autonomous vehicle, wherein identifying the exception handler includes;

 

in response to a determination that the set of available handlers includes an internal exception handler that models the exceptional condition, identifying the internal exception handler as the active handler; and

 

in response to a determination that the set of available handlers omits the internal exception handler that models the exceptional condition, identifying an external assistance exception handler as the active handler;

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an unexceptional condition, identifying a primary handler from the set of available handlers as the active handler for controlling the autonomous vehicle;

controlling the autonomous vehicle to traverse a current portion of the vehicle transportation network in accordance with the active handler; and

in response to receiving an indicator from the active handler subsequent to controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the active handler;

generating an updated exception belief state probability distribution based on the current exception belief state probability distribution and the indicator; and

traversing a subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution.

11. The autonomous vehicle of claim 10, wherein the processor is configured to execute the instructions stored on the non-transitory computer readable medium to control the autonomous vehicle to traverse the vehicle transportation network, wherein:

operating the introspective autonomous vehicle operational management controller includes the introspective autonomous vehicle operational management controller;

in response to a determination that the current exception belief state probability distribution of the introspective autonomous vehicle operational management policy indicates an uncertainty value that exceeds a defined uncertainty threshold;

identifying a vehicle control action for traversing the current portion of the vehicle transportation network;

controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the vehicle control action;

in response to controlling the autonomous vehicle to traverse the current portion of the vehicle transportation network in accordance with the vehicle control action, generating an updated exception belief state probability distribution that indicates an uncertainty value that is within the defined uncertainty threshold; and

traversing a subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution.

12. The autonomous vehicle of claim 10, wherein the primary handler includes a model of a defined limited autonomous vehicle operational management domain.

13. The autonomous vehicle of claim 12, wherein the defined limited autonomous vehicle operational management domain is a route planning domain and the model from the primary handler is a route planning model of the route planning domain.

14. The autonomous vehicle of claim 13, wherein the primary handler includes a scenario-specific operational control evaluation model of a distinct vehicle operational scenario.

15. The autonomous vehicle of claim 12, wherein the processor is configured to execute the instructions stored on the non-transitory computer readable medium to control the autonomous vehicle to traverse the vehicle transportation network, wherein:

the set of exceptions includes one or more exceptional factors that describe exceptional conditions that probabilistically affect the operation of the autonomous vehicle, wherein the model of the defined limited autonomous vehicle operational management domain omits a description of the exceptional conditions.

16. The autonomous vehicle of claim 10, wherein the introspective autonomous vehicle operational management model includes a defined set of indicators, wherein the defined set of indicators includes the indicator, and wherein:

identifying the primary handler from the set of available handlers as the active handler includes identifying the primary handler from the set of available handlers as the active handler such that the primary handler outputs the indicator from the defined set of indicators in accordance with a defined temporal term;

identifying the exception handler from the set of available handlers as the active handler includes identifying the exception handler from the set of available handlers as the active handler such that the exception handler outputs the indicator from the defined set of indicators in response to identifying a termination condition, wherein the indicator is;

a success indicator indicating that the exceptional condition is resolved;

or

on a condition that the exception handler is the internal exception handler, a failure indicator indicating that a set of available actions available to the exception handler omits an available action for resolving the exceptional condition.

17. The autonomous vehicle of claim 16, wherein the introspective autonomous vehicle operational management model includes:

a defined set of introspective states based on a defined set of states and the defined set of exceptions;

a defined set of introspective actions based on a defined set of actions and the defined set of available handlers;

a transition profile representing a probability of a state from the defined set of states subsequent to traversing the vehicle transportation network from a current exceptional state from the defined set of introspective states in accordance with the active handler;

an exception profile representing a probability of an exception from the set of exceptions subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

a cost profile representing an expected immediate cost subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

an indicator profile representing a probability of observing an indicator from the defined set of indicators subsequent to traversing the vehicle transportation network from the current exceptional state from the defined set of introspective states in accordance with the active handler;

an introspective transition function based on the transition profile and the exception profile;

an introspective reward function based on the cost profile and the defined set of available handlers;

a set of introspective observations based on a defined set of observations and the defined set of indicators;

an introspective observation function based on the indicator profile and the defined set of available handlers; and

a defined set of target states.

18. The autonomous vehicle of claim 10, wherein the processor is configured to execute the instructions stored on the non-transitory computer readable medium to control the autonomous vehicle to traverse the subsequent portion of the vehicle transportation network based on the updated exception belief state probability distribution by:

on a condition that the active handler is the primary handler and in response to a determination that the updated exception belief state probability distribution indicates an exceptional condition;

suspending the primary handler; and

identifying an exception handler from the set of available handlers as the active handler.

19. A non-transitory computer-readable storage medium, comprising executable instructions that, when executed by a processor, facilitate performance of introspective autonomous vehicle operational management, comprising:

traversing, by an autonomous vehicle, a vehicle transportation network, wherein traversing the vehicle transportation network includes;

operating an introspective autonomous vehicle operational management controller, wherein the introspective autonomous vehicle operational management controller includes an introspective autonomous vehicle operational management policy for an introspective autonomous vehicle operational management model, wherein the introspective autonomous vehicle operational management model defines a set of available handlers, wherein the defined set of available handlers includes;

a primary hander that includes a model of a defined limited autonomous vehicle operational management domain that omits a description of one or more exceptional conditions; and

an internal exception handler that includes a model that includes a description of at least one exceptional condition from the one or more exceptional conditions;

wherein operating the introspective autonomous vehicle operational management controller includes;

identifying the primary handler as an active handler for controlling the autonomous vehicle to traverse a first portion of the vehicle transportation network for a defined temporal term;

in response to receiving an indicator from the primary handler subsequent to the autonomous vehicle to traversing the current portion of the vehicle transportation network in accordance with the primary handler for the defined temporal term, generating an exception belief state probability distribution;

in response to a determination that the exception belief state probability distribution indicates that a current state of the autonomous vehicle corresponds to the at least one exceptional condition;

suspending the primary handler;

identifying the internal exception handler as the active handler for controlling the autonomous vehicle to traverse a second portion of the vehicle transportation network; and

controlling the autonomous vehicle to traverse the second portion of the vehicle transportation network in accordance with the exception handler.

20. The non-transitory computer-readable storage medium of claim 19, wherein operating the introspective autonomous vehicle operational management controller includes:

in response to a determination that the exception belief state probability distribution indicates that the current state of the autonomous vehicle corresponds to an exceptional condition from the one or more exceptional conditions other than the at least one exceptional condition;

suspending the primary handler;

identifying an external assistance exception handler as the active handler for controlling the autonomous vehicle to traverse the second portion of the vehicle transportation network;

in response to receiving an indicator from the external assistance exception handler subsequent to the autonomous vehicle to traversing the current portion of the vehicle transportation network in accordance with the external assistance exception handler, generating an updated exception belief state probability distribution;

resuming the primary handler; and

controlling the autonomous vehicle to traverse a subsequent portion of the vehicle transportation network in response to the primary handler.