A Bisimulation-Based Approach to the Analysis of Human-Computer - - PowerPoint PPT Presentation

A Bisimulation-Based Approach to the Analysis

• f Human-Computer Interaction

S´ ebastien Comb´ efis Charles Pecheur

Universit´ e catholique de Louvain (UCLouvain) Belgium

July 16, 2009

[EICS’09, Pittsburgh, PA, USA]

Complex Systems

With Automation and Human-Interaction ◮ Accidents: bad system design, bad operator, wrong interaction

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 2

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model I n t e r a c t i

• n
• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model I n t e r a c t i

• n

Interacts

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model Abstracts

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model Induces

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model

Objective: Generate an abstraction of a given system model Motivation: Build training material to enforce a good mental model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Human-Computer Interaction

Different Components user manual, training system model system interface user mental model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 3

Outline

1 Introduction 2 Modelling Human-Computer Interaction 3 Generating Full-Control Mental Model 4 Conclusion

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 4

Models of the System

The Big Picture System Model Full-Control Model Operational Model Mental Model Environment Tasks Training Manual Social Learning . . .

derives influences

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 5

Models of the System

The Big Picture System Model Full-Control Model Operational Model Mental Model Environment Tasks Training Manual Social Learning . . .

derives influences ALL BEHAVIOUR

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 5

Models of the System

The Big Picture System Model Full-Control Model Operational Model Mental Model Environment Tasks Training Manual Social Learning . . .

derives influences PARTIAL BEHAVIOUR

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 5

Models of the System

The Big Picture System Model Full-Control Model Operational Model Mental Model Environment Tasks Training Manual Social Learning . . .

derives influences CHANGES

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 5

Labelled Transition Systems

The Vehicle Transmission System Example ◮ Semi-automatic gearbox (Degani, 2007)

GEAR LEVER

push-up pull-down

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 6

Labelled Transition Systems

The Vehicle Transmission System Example ◮ System modelled as a Labelled Transition System (LTS)

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

push-up pull-down up down state transition with action

LTS executions yield traces Action-Based Interface: command , observation , τ

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 6

Vehicle Transmission System

A Mental Model ◮ The user sees the system as a three-state system

high medium low

push-up pull-down up down

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Vehicle Transmission System

A Mental Model ◮ The user sees the system as a three-state system

high medium low

push-up pull-down up down

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Capturing Possible Interactions

Synchronous Parallel Composition

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3 high medium low

push-up pull-down up down

System Model Mental Model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Capturing Possible Interactions

Synchronous Parallel Composition

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3 high medium low

push-up pull-down up down

System Model Mental Model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Capturing Possible Interactions

Synchronous Parallel Composition

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3 high medium low

push-up pull-down up down

System Model Mental Model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Capturing Possible Interactions

Synchronous Parallel Composition

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3 high medium low

push-up pull-down up down

System Model Mental Model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 7

Full-Control Mental Model

Capturing All Behaviours of the System Definition (Full-control mental model) A mental model allows full-control of a system iff for all sequences of observable actions σ such that s0M

σ

= = ⇒ sM and s0U

σ

− − → sU: Ac(sM) = Ac(sU) ∧ Ao(sM) ⊆ Ao(sU) ◮ Intuition: For each state in the synchronous parallel composition: Exactly same commands on system and mental models At least all observations of system model on mental model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 8

Outline

1 Introduction 2 Modelling Human-Computer Interaction 3 Generating Full-Control Mental Model 4 Conclusion

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 9

Mental Model Generation

Full-Control Equivalence Generating a minimal full-control mental model for a system Defining an equivalence relation ≈fc on system’s states Merging equivalent states together to get a reduced model

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 10

Mental Model Generation

Full-Control Equivalence Generating a minimal full-control mental model for a system Defining an equivalence relation ≈fc on system’s states Merging equivalent states together to get a reduced model s ≈fc t if and only if

s t

≈fc

s’

α ≈fc

t’

α ◮ α a command

∀s

α

= = ⇒ s′ ∃t

α

= = ⇒ t′ : s′ ≈fc t′

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 10

Mental Model Generation

Full-Control Equivalence Generating a minimal full-control mental model for a system Defining an equivalence relation ≈fc on system’s states Merging equivalent states together to get a reduced model s ≈fc t if and only if

s t

≈fc

s’

β ◮ α a command

∀s

α

= = ⇒ s′ ∃t

α

= = ⇒ t′ : s′ ≈fc t′

◮ β an observation

∀s

β

= = ⇒ s′ : ∃t

β

= = ⇒ t′ ⇒ s′ ≈fc t′

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 10

Mental Model Generation

Full-Control Equivalence Generating a minimal full-control mental model for a system Defining an equivalence relation ≈fc on system’s states Merging equivalent states together to get a reduced model s ≈fc t if and only if

s t

≈fc

s’

β ≈fc

t’

β ◮ α a command

∀s

α

= = ⇒ s′ ∃t

α

= = ⇒ t′ : s′ ≈fc t′

◮ β an observation

∀s

β

= = ⇒ s′ : ∃t

β

= = ⇒ t′ ⇒ s′ ≈fc t′

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 10

Mental Model Generation

Full-Control Equivalence Generating a minimal full-control mental model for a system Defining an equivalence relation ≈fc on system’s states Merging equivalent states together to get a reduced model s ≈fc t if and only if

s t

≈fc

s’

ε ≈fc

t’

ε ◮ α a command

∀s

α

= = ⇒ s′ ∃t

α

= = ⇒ t′ : s′ ≈fc t′

◮ β an observation

∀s

β

= = ⇒ s′ : ∃t

β

= = ⇒ t′ ⇒ s′ ≈fc t′

◮ ε an empty trace

∀s

ε

= ⇒ s′ ∃t

ε

= ⇒ t′ : s′ ≈fc t′

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 10

Full-Control Equivalence

fc-determinism Definition (fc-determinism) A model is fc-deterministic iff for all traces σ (including ε) such that s0

σ

= = ⇒ s and s0

σ

= = ⇒ s′, we have s ≈fc s′. α α β γ α τ β

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 11

Full-Control Equivalence

fc-determinism Definition (fc-determinism) A model is fc-deterministic iff for all traces σ (including ε) such that s0

σ

= = ⇒ s and s0

σ

= = ⇒ s′, we have s ≈fc s′. α α β γ α τ β Theorem If a system model is fc-deterministic, then minimizing it wrt. ≈fc gives a mental model that allows full-control of the system

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 11

Mental Model Generation

Step I: Compute the Equivalence

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

◮ Reduction done with a variant of Paige-Tarjan algorithm

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Mental Model Generation

Step I: Compute the Equivalence

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Mental Model Generation

Step I: Compute the Equivalence

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Mental Model Generation

Step I: Compute the Equivalence

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Mental Model Generation

Step I: Compute the Equivalence

high-1 high-2 high-3 medium-1 medium-2 low-1 low-2 low-3

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Mental Model Generation

Step II: Reduce the System

high medium low-a low-b low-c

◮ When in LOW mode, the user must track up and down actions

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 12

Conclusion

Contributions Definition of the full-control property and equivalence Generation of a minimal full-control mental model Implementation of a prototype

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 13

Conclusion

Perspectives Generating mode-preserving mental model Applying developped techniques on real-world example Considering state-based interface Generating minimal operational mental model Modelling “imperfect users” (user errors)

• S. Comb´

efis, C. Pecheur (UCLouvain) – A Bisimulation-Based Approach to the Analysis of HCI 14