IMPORTANT, MUST BE READ... : SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation
Title : SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation
SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation
IMPORTANT, MUST BE READ... Challenges inwards Autonomous Vehicle Testing together with Validation from Philip Koopman
Developers of autonomous systems appear upwards distinct challenges inwards conforming to established methods of validating safety. It is good known that testing lonely is insufficient to assure safety, because testing long plenty to found ultra-dependability is to a greater extent than oft than non impractical. That’s why software security standards emphasize high character evolution processes. Testing together with thence validates procedure execution rather than conduct validating dependability.
Two pregnant challenges arise inwards applying traditional security processes to autonomous vehicles. First, precisely gathering a consummate ready of organization requirements is hard because of the sheer number of combinations of possible scenarios together with faults. Second, autonomy systems commonly usage machine learning (ML) inwards a agency that makes the requirements together with pattern of the organization opaque. After training, normally nosotros know what an ML element volition practise for an input it has seen, but to a greater extent than oft than non not what it volition practise for at to the lowest degree another inputs until nosotros attempt them. Both of these issues arrive hard to line requirements together with designs to testing every bit is required for executing a security validation process. In other words, we’re edifice systems that can’t move validated due to incomplete or fifty-fifty unknown requirements together with designs.
Adaptation makes the occupation fifty-fifty worse past times making the organization that must move validated a moving target. In the full general case, it is impractical to validate all the possible adaptation states of an autonomy organization using traditional security pattern processes.
An approach that tin assistance amongst the requirements, design, together with adaptation problems is basing a security declaration non on correctness of the autonomy functionality itself, but rather on conformance to a ready of security envelopes. Each security envelope describes a boundary inside the operational Blue Planet infinite of the autonomy system.
Influenza A virus subtype H5N1 organization operating inside a “safe” envelope knows that it’s prophylactic together with tin operate amongst total autonomy. Influenza A virus subtype H5N1 organization operating inside an “unsafe” envelope knows that it’s unsafe, together with must invoke a failsafe action. Multiple partial specifications tin move used every bit an envelope set, amongst the intersection of prophylactic envelopes permitting total autonomy, together with the spousal human relationship of dangerous envelopes provoking validated, together with potentially complex, failsafe responses.
Envelope mechanisms tin move implemented using traditional software technology scientific discipline techniques, reducing the problems amongst requirements, design, together with adaptation that would otherwise impede security validation. Rather than attempting to bear witness that autonomy volition ever piece of work correctly (which is nevertheless a valuable finish to ameliorate availability), the envelope approach measures the demeanor of i or to a greater extent than autonomous components to create upwards one's hear if the consequence is safe. While this is non necessarily an tardily affair to do, at that spot is argue to believe that checking autonomy behaviors for security is easier than implementing perfect, optimized autonomy actions. This envelope approach mightiness move used to observe faults during evolution together with to trigger failsafes inwards fleet vehicles.
Inevitably at that spot volition move tension betwixt simplicity of the envelope definitions together with permissiveness, amongst to a greater extent than permissive envelope definitions probable beingness to a greater extent than complex. Operating inwards the gap areas betwixt “safe” together with “unsafe” requires human supervision, because the autonomy organization can’t move certain it is safe.
One agency to await at the progression from partial to total autonomy is that, over time, systems tin growth permissiveness past times defining together with growing “safe” envelopes, shrinking “unsafe” envelopes, together with eliminating whatever gap areas.
ACM Reference format:
P. Koopman, 2017. Challenges inwards Autonomous Vehicle Validation. In
Proceedings of 1st International Workshop on Safe Control of Connected
together with Autonomous Vehicles, Pittsburgh, Pennsylvania, USA, Apr 2017
(SCAV 2017), 1 page.
Permission to brand digital or hard copies of constituent or all of this piece of work for personal or classroom usage is granted without fee provided that copies are non made or distributed for turn a profit or commercial payoff together with that copies comport this notice together with the total citation on the start page. Copyrights for third-party components of this piece of work must move honored. For all other uses, contact the Owner/Author.
Copyright is held past times the owner/author(s).
SCAV'17, Apr 21-21 2017, Pittsburgh, PA, USA
ACM 978-1-4503-4976-5/17/04.
http://dx.doi.org/10.1145/3055378.3055379
Challenges inwards Autonomous Vehicle Validation
Keynote Presentation Abstract
Philip Koopman
Carnegie Mellon University; Edge Case Research LLC
ECE Dept. HH A-308, 5000 Forbes Ave., Pittsburgh, PA, USA
koopman@cmu.edu
Developers of autonomous systems appear upwards distinct challenges inwards conforming to established methods of validating safety. It is good known that testing lonely is insufficient to assure safety, because testing long plenty to found ultra-dependability is to a greater extent than oft than non impractical. That’s why software security standards emphasize high character evolution processes. Testing together with thence validates procedure execution rather than conduct validating dependability.
Two pregnant challenges arise inwards applying traditional security processes to autonomous vehicles. First, precisely gathering a consummate ready of organization requirements is hard because of the sheer number of combinations of possible scenarios together with faults. Second, autonomy systems commonly usage machine learning (ML) inwards a agency that makes the requirements together with pattern of the organization opaque. After training, normally nosotros know what an ML element volition practise for an input it has seen, but to a greater extent than oft than non not what it volition practise for at to the lowest degree another inputs until nosotros attempt them. Both of these issues arrive hard to line requirements together with designs to testing every bit is required for executing a security validation process. In other words, we’re edifice systems that can’t move validated due to incomplete or fifty-fifty unknown requirements together with designs.
Adaptation makes the occupation fifty-fifty worse past times making the organization that must move validated a moving target. In the full general case, it is impractical to validate all the possible adaptation states of an autonomy organization using traditional security pattern processes.
An approach that tin assistance amongst the requirements, design, together with adaptation problems is basing a security declaration non on correctness of the autonomy functionality itself, but rather on conformance to a ready of security envelopes. Each security envelope describes a boundary inside the operational Blue Planet infinite of the autonomy system.
Influenza A virus subtype H5N1 organization operating inside a “safe” envelope knows that it’s prophylactic together with tin operate amongst total autonomy. Influenza A virus subtype H5N1 organization operating inside an “unsafe” envelope knows that it’s unsafe, together with must invoke a failsafe action. Multiple partial specifications tin move used every bit an envelope set, amongst the intersection of prophylactic envelopes permitting total autonomy, together with the spousal human relationship of dangerous envelopes provoking validated, together with potentially complex, failsafe responses.
Envelope mechanisms tin move implemented using traditional software technology scientific discipline techniques, reducing the problems amongst requirements, design, together with adaptation that would otherwise impede security validation. Rather than attempting to bear witness that autonomy volition ever piece of work correctly (which is nevertheless a valuable finish to ameliorate availability), the envelope approach measures the demeanor of i or to a greater extent than autonomous components to create upwards one's hear if the consequence is safe. While this is non necessarily an tardily affair to do, at that spot is argue to believe that checking autonomy behaviors for security is easier than implementing perfect, optimized autonomy actions. This envelope approach mightiness move used to observe faults during evolution together with to trigger failsafes inwards fleet vehicles.
Inevitably at that spot volition move tension betwixt simplicity of the envelope definitions together with permissiveness, amongst to a greater extent than permissive envelope definitions probable beingness to a greater extent than complex. Operating inwards the gap areas betwixt “safe” together with “unsafe” requires human supervision, because the autonomy organization can’t move certain it is safe.
One agency to await at the progression from partial to total autonomy is that, over time, systems tin growth permissiveness past times defining together with growing “safe” envelopes, shrinking “unsafe” envelopes, together with eliminating whatever gap areas.
ACM Reference format:
P. Koopman, 2017. Challenges inwards Autonomous Vehicle Validation. In
Proceedings of 1st International Workshop on Safe Control of Connected
together with Autonomous Vehicles, Pittsburgh, Pennsylvania, USA, Apr 2017
(SCAV 2017), 1 page.
Permission to brand digital or hard copies of constituent or all of this piece of work for personal or classroom usage is granted without fee provided that copies are non made or distributed for turn a profit or commercial payoff together with that copies comport this notice together with the total citation on the start page. Copyrights for third-party components of this piece of work must move honored. For all other uses, contact the Owner/Author.
Copyright is held past times the owner/author(s).
SCAV'17, Apr 21-21 2017, Pittsburgh, PA, USA
ACM 978-1-4503-4976-5/17/04.
http://dx.doi.org/10.1145/3055378.3055379
IMPORTANT, MUST BE READ...
Thank for your attention SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation
my blog SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation, Have a nice day.
Now you read article SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation this permalink article is http://fairemirima.blogspot.com/2017/11/scav-2017-keynote-challenges-inward.html Thank you and Best regards. You Can read nice Tips below. It was always better to choose topics that interest you or in wich you at least have some knowledge about . When creating targeted internet copywriting , you have to stick with your strong points , or everyone will know it . Make a list of all of the things and or topics that you are interested in . . . How much do you know ? Can you tell it as a story ? That is The essence of writing for the web . You Have to know your subject well , or nobody will believe you it is always better to impress someone then upset them . When Writing Targeted Internet Copywriting , you have to choose your appropriate target group of customers . without a target group of customers , you could ramble on incessantly about random subjects for days on end with no essence of a final goal . You always have to keep in mind who your customers are and what they are looking for . . . . . . . . . IMPORTANT, MUST BE READ...
0 Response to "SCAV 2017 Keynote: Challenges inward Autonomous Vehicle Validation"
Post a Comment