Towards sustainable safety and security in autonomous ...

24
University of Luxembourg SnT - CritiX Towards sustainable safety and security in autonomous vehicles Marcus Völp [email protected]

Transcript of Towards sustainable safety and security in autonomous ...

University of LuxembourgSnT - CritiX

Towards sustainable safety and security in autonomous vehicles

Marcus Völp

[email protected]

Autonomous driving – the next complexity milestone

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 2

Source: http://paleofuture.com/blog/2010/12/9/driverless-car-of-the-future-1957.html

Level 4 autonomy => Level 5

Source: Wiki SelKet (CC-BY-SA-3.0)

Brain areas involved in human visual perception

Recognition not only of regular traffic

Ethics

Source: Autobild

Acceptance and Reputation

Implicit Communication

1957

Presenter
Presentation Notes
62 years Safe and secure fully autonomous driving will be the next complexity milestone we have to master … - Dorsal (green) and ventral (purple) stream of visual cortex (14-16 billion neurons) – 86 billion total Psychological scientists at Université de Bretagne-Sud:  45% vs. 68% of drivers stopped - Instead of making vision as perfect, or ideally better than the human, we could of course also make all traffic participants sentient, having them share their position with nearby cars, etc. => this is the other use of yellow vests.

Complexity of autonomous driving:

• Level 3: 300 MLOC (human supervision)

• Level 5: 1 BLOC+ ?

Autonomous driving – the next complexity milestone

Current Cars:

• ~ 100 MLOC (30 MLOC multimedia)

• ~ 100 ECUs

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 3

Components associated with physical control of the vehicle

Components associated with safety

Components associated with entertainment and convenience

Image credit: Mercedes-Benz Museum (as cited in Computer

History Museum, 2011)Slide from Intel ADG

Functionality vs. Complexity

Complexity of autonomous driving:

• Level 3: 300 MLOC (human supervision)

• Level 5: 1 BLOC+ ?

Autonomous driving – the next complexity milestone

Current Cars:

• ~ 100 MLOC (30 MLOC multimedia)

• ~ 100 ECUs with 25 – 200 microprocessors

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 4

Components associated with physical control of the vehicle

Components associated with safety

Components associated with entertainment and convenience

Image credit: Mercedes-Benz Museum (as cited in Computer

History Museum, 2011)Slide from Intel ADG

Functionality vs. Complexity

Over the air updates

https://arstechnica.com/cars/2017/07/gm-to-offer-ota-software-updates-before-2020-but-only-for-a-new-infotainment-platform/

Safety Certification?

Complexity of autonomous driving:

• Level 3: 300 MLOC (human supervision)

• Level 5: 1 BLOC+ ?

Autonomous driving – the next complexity milestone

Current Cars:

• ~ 100 MLOC (30 MLOC multimedia)

• ~ 100 ECUs with 25 – 200 microprocessors

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 5

Components associated with physical control of the vehicle

Components associated with safety

Components associated with entertainment and convenience

Image credit: Mercedes-Benz Museum (as cited in Computer

History Museum, 2011)Slide from Intel ADG

Functionality vs. Complexity

Over the air updates

https://arstechnica.com/cars/2017/07/gm-to-offer-ota-software-updates-before-2020-but-only-for-a-new-infotainment-platform/

Safety Certification?

We need: 1. fault and intrusion tolerance for safety critical / real-time systems2. safety and security for the entire lifetime of the car

(avg. 11.6 year on US roads [IHS Markit Report ‘18])3. automatic resilience

Presenter
Presentation Notes
Paulo: no midget security officer in the trunk (even Tesla‘s fleet control doesn‘t scale)

Threat Vectors on autonomous and cooperativevehicle ecosystems

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 6

I2I

V2V

Internet

I2I

ECUV2I

I2I

I2I

1

5

1

2 3

48

7

6

T.A.

RSU System Provider

Threat Vectors:1. Attacks on global V2I/I2I

communication infrastructure2. Attacks on local V2V

communication infrastructure3. Attacks on in-vehicle

communication infrastructure4. Attacks on vehicle computing

nodes' software5. Attacks on road-side

units'software6. Attacks on sensors and

control-sensitive data7. Attacks on authentication

mechanisms8. Physical-level attacks

[Lima et al. “Towards Safe and Secure Autonomous and Cooperative Vehicle Ecosystems”, CPS-SPC 2016]

Outline

Motivation

What we all know: FIT / Resilience

Full compromise of swarm individuals is intolerable

Intra Vehicular Systems

Towards Sustainable Safety and Security Surviving perception / maneuver planning unavailabilities Lifecycle for safeguarding safety and security

ConclusionsMarcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 7

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 8

NodeIncomingTraffic

Client

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 9

NodeIncomingTraffic

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 10

NodeIncomingTraffic

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 11

NodeIncomingTraffic

Intrusion Prevention

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 12

NodeIncomingTraffic

Intrusion Prevention

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

P. Sousa et al. – Exhaustion Failure

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 13

NodeIncomingTraffic

Intrusion Prevention

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

P. Sousa et al. – Exhaustion Failure

Fault and Intrusion Tolerance

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 14

NodeIncomingTraffic

Intrusion Prevention

Node

Node

Consensus(in fault-threshold exceeding quorum)

Clientf+1

P. Sousa et al. – Exhaustion Failure

Presenter
Presentation Notes
Nodes can become fully compromised (or, if we talk about hybrid protocols such as MinBFT, at least up to a very simple trusted-trustworthy component)

Full compromise of swarm individuals is intolerable

Platoon of Cars

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 15

distance keeping

driving

Platoon votes on common operation (e.g., maneuver to follow)

leader

Full compromise of swarm individuals is intolerable

Platoon of Cars

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 16

distance keeping

driving

Platoon votes on common operation (e.g., maneuver to follow)

leader

Presenter
Presentation Notes
If by doing so, we detect a rogue car, we can of course try and avoid them, but unlike in the cyber-only case, we are still left with a fully compromised node that can cause physical harm

Full compromise of swarm individuals is intolerable

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 17

Presenter
Presentation Notes
Bosch / Conti / Denso

Towards Sustainable Safety and Security

Safeguard safety through trusted trustworthy components

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 18

distance keeping

driving

Platoon votes on common operation (e.g., maneuver to follow)

leader

A

Towards Sustainable Safety and Security

Known Strategies for TCB Reduction

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 19

B

C

B

C

D

e.g., C. Weinhold: JVPFS(also Inktag, SGX)

D

qed.

B1

C

DD

B2B3

split applications: trusted / untrusted reuse untrusted replication (trust majority; not individual)

Presenter
Presentation Notes
Abstractly: functionality

Towards Sustainable Safety and Security

Reusing potentially unavailable maneuver planning

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 20

fail save (A)

point of no return

A

Towards Sustainable Safety and Security

Reusing potentially unavailable maneuver planning

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 21

fail save (A)

point of no return

A

Towards Sustainable Safety and Security

Reusing potentially unavailable maneuver planning

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 22

fail save (A)

point of no return

A B

fail save (B)

Towards Sustainable Safety and Security

Lifecycle for safeguarding safety and security

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 23

timeend of lifeout of maintenanceout of productionin productiondesign

vulnerability

accidentalfault

fault

attack

maliciousfault

error failure

subsystem

Reactive andproactive recoveryand diversification

vulnerability removal(testing, verification,

…) fault diagnosis / removal

intrusion detection

S3-aware design

fault tolerance; error masking(e.g., through replication);

error detection andinitiation of recovery

replacement offaulty hardwarein new systems

patches / update of SW components

(active supply of div. pool)

criticalupdates

only

refill of diversification pool onlythrough on-board mechanisms

/ cross fleet fertilization

…secret cleaning

beforedecommissioning

Presenter
Presentation Notes
Make all effort to bring hybrids to zero defect: hybrid safeguards safety

This Talk in one Slide

Marcus Völp - Towards sustainable safety and security in autonomous vehicles - 75th IFIP WG 10.4 on Dependable Computing and Fault Tolerance - 2019 24

We are hiring bright PhD students and postdocs!