I A Visual Logic for the Description of Highway Traffic Scenarios

Stephanie Kemper, Christoph Etzien{kemper,etzien}@offis.de

OFFIS Institute for Information TechnologyR&D Division Transportation

December 6, 2013

CSD&M2013, Paris, France

I 1 Motivation

I Logic developed in context of advanced driver assistance system (ADAS)

development

I Specify driving manoeuvres as (sequences of) traffic situations on the highway

⇒ Relations and interactions between vehicles/vehicle components

I Interface between developers and scientists

I Simple and intuitive, easy to learn and use

I Formal foundation and semantics for further analysis

⇒ Bridge terminology gap between different domain experts

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 2 The Visual Logic (VL)Atoms

I Basic building blocks of the VL

I Single atom: traffic “snapshot”

I Sequence of atoms: traffic scenario (with evolution)

I Two parts

I Traffic View: spatial relations of vehicles

I Communication Description: communications/synchronisations during situation

sketched in Traffic View

I Compositionality of Atoms: if both Traffic View and Communication Description

are composable (later)

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 3 The Visual Logic (VL)Traffic View: Basics

I Describe spatial relations of vehicles

I Set of lane separators spans “canvas”

I Traffic flow from left to right, orientational notions accordingly

I Position defined via vehicle sides

I Separate position relations for two dimensions

ls1

ls2

ls3

lateral

longitudinal

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 3 The Visual Logic (VL)Traffic View: Basics

I Describe spatial relations of vehicles

I Set of lane separators spans “canvas”

I Traffic flow from left to right, orientational notions accordingly

I Position defined via vehicle sides

I Separate position relations for two dimensions

ls1

ls2

ls3

lateral

longitudinal

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 3 The Visual Logic (VL)Traffic View: Basics

I Describe spatial relations of vehicles

I Set of lane separators spans “canvas”

I Traffic flow from left to right, orientational notions accordingly

I Position defined via vehicle sides

I Separate position relations for two dimensions

ls1

ls2

ls3

lateral

longitudinal

left

right

back front

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 3 The Visual Logic (VL)Traffic View: Basics

I Describe spatial relations of vehicles

I Set of lane separators spans “canvas”

I Traffic flow from left to right, orientational notions accordingly

I Position defined via vehicle sides

I Separate position relations for two dimensions

ls1

ls2

ls3

lateral

longitudinal

B.l

B.r

B.b B.f

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 4 The Visual Logic (VL)Traffic View: Relations

I Longitudinal relations (binary): before (e.g. , ), meets ( , ), overlaps

( , ) (or respective inverse) or equals ( , ) for each pair of vehicles

⇒ induces total order on front/back vehicle sides

I Lateral relations: binary as before, plus unary (high-level) on_lane (e.g. ) or

between_lanes ( ) for each vehicle

⇒ induces total order on left/right vehicle sides and lane separators

I Distance arrows to further restrict relative positions (default: arrow with label

[0,∞[ to next element—not shown in pictures)

ls1

ls2

ls3

[50,∞[

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 4 The Visual Logic (VL)Traffic View: Relations

I Longitudinal relations (binary): before (e.g. , ), meets ( , ), overlaps

( , ) (or respective inverse) or equals ( , ) for each pair of vehicles

⇒ induces total order on front/back vehicle sides

I Lateral relations: binary as before, plus unary (high-level) on_lane (e.g. ) or

between_lanes ( ) for each vehicle

⇒ induces total order on left/right vehicle sides and lane separators

I Distance arrows to further restrict relative positions (default: arrow with label

[0,∞[ to next element—not shown in pictures)

ls1

ls2

ls3

[50,∞[

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 5 The Visual Logic (VL)Traffic View: Constraints

I Each Traffic View translated into spatial constraint

ls1

ls2

ls3

[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

I Implicit assumptions

I (B.l > B.r) ∧ (B.f > B.b) ∧ (R.l > R.r) ∧ ...

I No other vehicles/lanes apart from those depicted

I Compositionality of Traffic Views: if position changes (spatial constraints) adhere

to “physically reasonable” behaviour (details in paper)

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 5 The Visual Logic (VL)Traffic View: Constraints

I Each Traffic View translated into spatial constraint

ls1

ls2

ls3

[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

I Implicit assumptions

I (B.l > B.r) ∧ (B.f > B.b) ∧ (R.l > R.r) ∧ ...

I No other vehicles/lanes apart from those depicted

I Compositionality of Traffic Views: if position changes (spatial constraints) adhere

to “physically reasonable” behaviour (details in paper)

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 6 The Visual Logic (VL)Traffic View: Composition (Lane Change Example)

ls1

ls2

ls3

[50,∞[

[50,∞[[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

sc2: (ls3 > B.l) ∧ (B.r > ls2) ∧ (R.l ≥ ls2) ∧ (ls2 ≥ R.r) ∧ (R.b - B.f > 50)

sc3: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls3 > R.l) ∧ (R.r > ls2) ∧ (R.b - B.f > 50)

I Spatial constraints must be disjoint to find “transition point”

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 6 The Visual Logic (VL)Traffic View: Composition (Lane Change Example)

ls1

ls2

ls3

[50,∞[[50,∞[

[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

sc2: (ls3 > B.l) ∧ (B.r > ls2) ∧ (R.l ≥ ls2) ∧ (ls2 ≥ R.r) ∧ (R.b - B.f > 50)

sc3: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls3 > R.l) ∧ (R.r > ls2) ∧ (R.b - B.f > 50)

I Spatial constraints must be disjoint to find “transition point”

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 6 The Visual Logic (VL)Traffic View: Composition (Lane Change Example)

ls1

ls2

ls3

[50,∞[[50,∞[

[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

sc2: (ls3 > B.l) ∧ (B.r > ls2) ∧ (R.l ≥ ls2) ∧ (ls2 ≥ R.r) ∧ (R.b - B.f > 50)

sc3: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls3 > R.l) ∧ (R.r > ls2) ∧ (R.b - B.f > 50)

I Spatial constraints must be disjoint to find “transition point”

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 6 The Visual Logic (VL)Traffic View: Composition (Lane Change Example)

ls1

ls2

ls3

[50,∞[[50,∞[

[50,∞[

sc1: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls2 > R.l) ∧ (R.r > ls1) ∧ (R.b - B.f > 50)

sc2: (ls3 > B.l) ∧ (B.r > ls2) ∧ (R.l ≥ ls2) ∧ (ls2 ≥ R.r) ∧ (R.b - B.f > 50)

sc3: (ls3 > B.l) ∧ (B.r > ls2) ∧ (ls3 > R.l) ∧ (R.r > ls2) ∧ (R.b - B.f > 50)

I Spatial constraints must be disjoint to find “transition point”

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 7 The Visual Logic (VL)Communication Description

I Specify communications/synchronisations during situation sketched in Traffic View

I Feature subset of Live Sequence Charts (extension of Message Sequence

Charts)⇒ translation into automaton representation

I Communicating instances: communicating parts of vehicles in Traffic View

look set

turn

look unset

turn back

I Compositionality of Communication Descriptions: if set of communicating

instances is identical

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 8 The Visual Logic (VL)Semantics

I For each Atom: annotate Communication Description with spatial constraint

I Translate “single” Communication Description into automaton representation

ls1

ls2

ls3

[50,∞[

Condition (“barrier synchronisation”) Invariant

look set

turn

sc1

sc1

[50,∞[

look

sc2

[50,∞[

unset

turn back

sc3

sc3

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 8 The Visual Logic (VL)Semantics

I For each Atom: annotate Communication Description with spatial constraint

I Translate “single” Communication Description into automaton representation

ls1

ls2

ls3

[50,∞[

Condition (“barrier synchronisation”) Invariant

look set

turn

sc1

sc1

[50,∞[

look

sc2

[50,∞[

unset

turn back

sc3

sc3

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 8 The Visual Logic (VL)Semantics

I For each Atom: annotate Communication Description with spatial constraint

I Translate “single” Communication Description into automaton representation

look set

turn

sc1

sc1

looksc

2unset

turn back

sc3

sc3

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 8 The Visual Logic (VL)Semantics

I For each Atom: annotate Communication Description with spatial constraint

I Translate “single” Communication Description into automaton representation

look set

turn

sc1

sc1

looksc

2unset

turn back

sc3

sc3

[sc1] [sc1]

[sc1] [sc1]

[sc2] [sc2]

[sc3] [sc3]

[sc3]

sc1

look

set

turn

sc2

look

sc3

unset

turn back

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 9 ApplicationObserver-based Verification

I Use automaton as observer to observe occurence of specified behaviour during

nomal system behaviour

I Example

I Goal: Check usability of ADAS interface

I VL specification to describe driver-ADAS interaction during driving manoeuvres

I System behaviour given as set of sample trajectories (driving simulator, human

test driver with prototypical implementation of ADAS)

I Test driver behaviour compliant with VL specification?

I Yes: Final implementation of prototype

I No: Find out why not (non-straightforward!), adapt either VL specification or

prototype, check again

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 10 Conclusion

I VL to describe highway traffic scenarios with communications

I Purely graphical specification of scenarios and manoeuvres

I Intuitive and easy to understand

I Yet with formal semantics

I VL as interface between different domain experts

I System engineers and traffic psychologists developing ADASs

I Scientists modelling, analysing and verifying behaviour of ADASs

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

I 11 Future Work

I Extend VL with

I Full set of LSC features (for Communication Descriptions), e.g.

I Desired or forbidden behaviour?

I Behaviour to be observed once or repeatedly?

I Partial observation: non-occurence or major system failure?

I Vehicle-specific properties like speed or acceleration

I Appearance and disappearance of lanes

I Parallel work on generalisation of VL

I General 2D plane,

I User-defined (spatial) relations,

I Straight or arched trajectories, ...

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013

Thank You!

Stephanie Kemper, Christoph Etzien A Visual Logic for Traffic Scenarios December 6, 2013