Assessment and response mechanism for autonomous systems
Various systems and methods for implementing an assessment and response mechanism for autonomous systems are described herein. An assessment and response system for an autonomous system is configured to access a realm classification of an event; determine a hazard score based on the realm classification, a severity metric, a likelihood metric, an urgency metric, and a confidence level metric; identify, based on the hazard score, a responsive action; and record details of the hazard score determination and the responsive action in a decision ledger.
1. An assessment and response system for an autonomous system comprising:
a processor subsystem; and
a memory device including instructions, which when executed by the processor subsystem, cause the processor subsystem to:
access a realm classification of an event occurring at a roadway location, wherein the realm classification classifies the event into one of a known domain or an unknown domain;
determine a hazard score for possible outcomes of the event based on the realm classification, a severity metric, a likelihood metric, an urgency metric, and a confidence level metric, wherein the severity metric is based on severity of the possible outcomes, wherein the likelihood metric is based on controllability of the possible outcomes, and wherein the urgency metric is based on speed of an autonomous response required from the autonomous system to perform a corresponding action for each of the possible outcomes;
identify, based on the hazard score, a responsive action of a safety system of the autonomous system, wherein the responsive action includes causing actuation of at least one of steering, braking, or stabilization of the autonomous system, and wherein the responsive action is identified based on: (i) whether the event is classified into the known domain or the unknown domain, (ii) the speed of the autonomous response required from the autonomous system to perform the responsive action, and (iii) whether the responsive action complies with one or more ethics rules applicable to the roadway location, wherein the one or more ethics rules are evaluated separately from rules of safety;
control the safety system of the autonomous system to perform the responsive action; and
record details of the hazard score determination and the responsive action in a decision ledger.
2. The system of claim 1 , wherein the known domain includes recognized events and the unknown domain includes unrecognized events.
3. The system of claim 1 , wherein the realm classification is performed by a realm classifier model that analyzes the event, and wherein the realm classifier model is to further classify the event as having one of an acceptable outcome or an unacceptable outcome.
4. The system of claim 1 , wherein the likelihood metric includes an exposure metric.
5. The system of claim 1 , wherein the likelihood metric includes a controllability metric.
6. The system of claim 1 , wherein to determine the hazard score, the processor subsystem is to determine the confidence level metric based on confidence in the severity metric, the likelihood metric, the urgency metric, and the possible outcomes of the event.
7. The system of claim 1 , wherein to determine the hazard score, the processor subsystem is to multiply the severity metric, the likelihood metric, the urgency metric, and the confidence level metric.
8. The system of claim 1 , wherein to identify the responsive action, the processor subsystem is to:
obtain a plurality of possible responsive actions involving causing at least one of steering, braking, or stabilization of the autonomous system;
access a policy bank to obtain the one or more ethics rules to evaluate the plurality of possible responsive actions; and
select the responsive action based on, at least in part, the one or more ethics rules from the policy bank.
9. The system of claim 1 , wherein to record the details of the hazard score determination and the responsive action in the decision ledger, the processor subsystem is to:
access a blockchain; and
write the details of the hazard score determination and the responsive action to the blockchain.
10. A method of providing an assessment and response system for an autonomous system, the method comprising:
accessing a realm classification of an event occurring at a roadway location, wherein the realm classification classifies the event into one of a known domain or an unknown domain;
determining a hazard score for possible outcomes of the event based on the realm classification, a severity metric, a likelihood metric, an urgency metric, and a confidence level metric, wherein the severity metric is based on severity of the possible outcomes, wherein the likelihood metric is based on controllability of the possible outcomes, and wherein the urgency metric is based on speed of an autonomous response required from the autonomous system to perform a corresponding action for each of the possible outcomes;
identifying, based on the hazard score, a responsive action of a safety system of the autonomous system, wherein the responsive action includes causing actuation of at least one of steering, braking, or stabilization of the autonomous system, and wherein the responsive action is identified based on: (i) whether the event is classified into the known domain or the unknown domain, (ii) the speed of the autonomous response required from the autonomous system to perform the responsive action, and (iii) whether the responsive action complies with one or more ethics rules applicable to the roadway location, wherein the one or more ethics rules are evaluated separately from rules of safety;
causing the safety system of the autonomous system to perform the responsive action; and
recording details of the hazard score determination and the responsive action in a decision ledger.
11. The method of claim 10 , wherein the known domain includes recognized events and the unknown domain includes unrecognized events.
12. The method of claim 10 , wherein the realm classification is performed by a realm classifier model that analyzes the event, and wherein the realm classifier model classifies the event as having one of an acceptable outcome or an unacceptable outcome.
13. The method of claim 10 , wherein the likelihood metric includes an exposure metric.
14. The system of claim 10 , wherein the likelihood metric includes a controllability metric.
15. At least one non-transitory machine-readable medium including instructions for providing an assessment and response system for an autonomous system, which when executed by a machine, cause the machine to:
access a realm classification of an event occurring at a roadway location, wherein the realm classification is performed by a realm classifier model that analyzes the event and classifies the event into one of a known domain or an unknown domain;
determine a hazard score for possible outcomes of the event based on the realm classification, a severity metric, a likelihood metric, an urgency metric, and a confidence level metric, wherein the severity metric is based on severity of the possible outcomes, wherein the likelihood metric is based on controllability of the possible outcomes, and wherein the urgency metric is based on speed of an autonomous response required from the autonomous system to perform a corresponding action for each of the possible outcomes;
identify, based on the hazard score, a responsive action of a safety system of the autonomous system, wherein the responsive action includes causing actuation of at least one of steering, braking, or stabilization of the autonomous system, and wherein the responsive action is identified based on:
(i) whether the event is classified into the known domain or the unknown domain, (ii) the speed of the autonomous response required from the autonomous system to perform the responsive action, and (iii) whether the responsive action complies with one or more ethics rules applicable to the roadway location, wherein the one or more ethics rules are evaluated separately from rules of safety;
cause the safety system of the autonomous system to perform the responsive action; and
record details of the hazard score determination and the responsive action in a decision ledger.
16. The at least one non-transitory machine-readable medium of claim 15 , wherein to determine the hazard score, the instructions cause the machine to determine the confidence level metric based on confidence in the severity metric, the likelihood metric, the urgency metric, and the possible outcomes of the event.
17. The at least one non-transitory machine-readable medium of claim 15 , wherein to determine the hazard score, the machine is to multiply the severity metric, the likelihood metric, the urgency metric, and the confidence level metric.
18. The at least one non-transitory machine-readable medium of claim 15 , wherein to identify the responsive action, the instructions cause the machine to:
obtain a plurality of possible responsive actions involving causing at least one of steering, braking, or stabilization of the autonomous system;
access a policy bank to obtain the one or more ethics rules to evaluate the plurality of possible responsive actions; and
select the responsive action based on, at least in part, the one or more ethics rules from the policy bank.
19. The at least one non-transitory machine-readable medium of claim 15 , wherein to record the details of the hazard score determination and the responsive action in the decision ledger, the instructions cause the machine to:
access a blockchain; and
write the details of the hazard score determination and the responsive action to the blockchain.