Zero Latency UI: Why We Care, How We’re Doing It 2013‐10‐16 1

2013‐10‐16
Zero Latency UI: Why We
Care, How We’re Doing It
1
2013‐10‐16
Sarah Williams
Paul Dietz
Maxim Mazeev
Robert Levy
Mike Cronin
Hrvoje Benko
Katie White
Albert Ng
Daniel Wigdor, Sarah Williams, Michael Cronin, Robert Levy, Katie
White, Maxim Mazeev, and Hrvoje Benko. 2009. Ripples: utilizing percontact visualizations to improve user interaction with touch
displays. In Proceedings of the 22nd annual ACM symposium on User
interface software and technology (UIST ’09). ACM, New York, NY, USA,
3-12.
Albert Ng, Julian Lepinski, Daniel Wigdor, Steven Sanders, and Paul
Dietz. 2012. Designing for low-latency direct-touch input. In
Proceedings of the 25th annual ACM symposium on User interface
software and technology (UIST ’12). ACM, New York, NY, USA, 453-464.
Ricardo Jota, Albert Ng, Paul Dietz, and Daniel Wigdor. 2013. How Fast
is Fast Enough? A Study of the Effects of Latency in Direct-Touch
Pointing Tasks. In Proceedings of the 2013 ACM annual conference on
Human Factors in Computing Systems (CHI ’13). ACM, New York, NY,
USA, (in press).
2
2013‐10‐16
High Performance Input
High Performance Input
Touch & Feedback
High Performance Input
High Performance Input
Touch & Feedback
Touch & Feedback
Perception of Latency
Perception of Latency
Performance under Latency
High Performance Input
High Performance Input
Touch & Feedback
Touch & Feedback
Perception of Latency
Perception of Latency
Performance under Latency
Performance under Latency
Architecting Performance
Architecting Performance
3
2013‐10‐16
HELP!
Raw Input
Error Ambiguity:
6 Causes
4
2013‐10‐16
Activation Action
Activation Action
Accidental Activation
Accidental Activation
Disabled Content
Disabled Content
5
2013‐10‐16
Fat Fingers
& Selection
Fat Fingers
& Selection
Manipulation Constraints
Manipulation Constraints
6
2013‐10‐16
Over-Capture
Over-Capture
Error Ambiguity Resolution
7
2013‐10‐16
Capture
Capture
Fat Fingers
Fat Fingers
Fat Fingers
Over-Capture
8
2013‐10‐16
Over-Capture
Tabletop Debris
Over-Capture
Tabletop Debris
Over-Capture Buttons
Over-Capture Buttons
Constraint Visualization
Constraint Visualization:
Resize
9
2013‐10‐16
Constraint Visualization:
Resize
Constraint Visualization:
Scroll
Constraint Visualization:
Scroll
Constraint Visualization:
Orthogonal
Constraint Visualization:
Orthogonal
10
2013‐10‐16
Study: Fat Fingers
Errors
15
With Feedback
Ripples
Without Feedback
no‐Ripples
10
5
0
33px
43px
53px
Study: Fat Fingers
2
8 3
Study: Preference
11
2013‐10‐16
I like to see what is happening, the visuals
help me figure out what's going on
I know where I am touching technically and I’m reassured
I know where I am touching technically and I’m reassured
it lets you know it was getting the signal
I know where I am touching technically and I’m reassured
the sensors caught my
actions better
[the] table was faster
it lets you know it was getting the signal
12
2013‐10‐16
[the] table was faster
This one was more responsive
Filmed at 80 fps
Avoiding Lag
Filmed at 80 fps
Avoiding Lag
13
2013‐10‐16
14
2013‐10‐16
15
2013‐10‐16
High Performance Input
Touch & Feedback
Perception of Latency
Performance under Latency
Architecting Performance
16
2013‐10‐16
Footage at 1/8 speed
Footage at 1/8 speed
17
2013‐10‐16
Sources of Direct-Touch Latency
Low Latency Touch Demo
CPU/GPU
Input
Sensor
Device
Driver
Operating
System
~5-17 ms
Application
~25-100 ms
Graphics
Display
Custom
Input
Sensor
~17 ms
1 ms
FPGA Processing
< 0.1 ms
High
Speed
Display
< 0.1 ms
Total Latency: 1 ms
Lag Distance @ 1 m/s: 0.1 cm
2
3
Touch perception
4
1
Perception response
Distance to
surface
Distance to
surface
Perception response
Time
Time
18
2013‐10‐16
Perception response
Distance to
surface
Distance to
surface
Perception response
Perception response
Perception response
Distance to
surface
Time
Distance to
surface
Time
first touch
Time
feedback
Time
Low Latency touch device
19
2013‐10‐16
First Experiment
High-speed processing
Just-noticeable difference (JND)
1 ms - 120 ms
75% accuracy rate
Low Latency touch device
20 participants
First Experiment
Latency
120ms
Results
0ms
Trial
Results
Latency (in ms)
Latency (in ms)
Results
20
2013‐10‐16
Latency (in ms)
Results
Dragging
24 ms
Condition A
Audience Test
60 ms
Condition B
Audience Test
42 ms
21
2013‐10‐16
Condition A
Condition B
Condition A
Audience Test
26 ms
Condition B
Audience Test
10 ms
22
2013‐10‐16
Condition A
Condition B
Condition A
Audience Test
5 ms
Condition B
Audience Test
2 ms
23
2013‐10‐16
Condition A
Condition B
Mean JND by Participant
Mean JND by Participant
Mean JND by Participant
High Performance Input
Touch & Feedback
Perception of Latency
Performance under Latency
2 ms
Architecting Performance
24
2013‐10‐16
Fitts’ Law
W
How does latency affect directtouch pointing tasks?
D
Fitts’ Law
Fitts’ Law
MacKenzie, I. S., and Ware, C. (1993). Lag as a determinant of human performance in interactive systems. ACM CHI 1993.
Target width
Target distance
+
2 cm
+
3 cm
4 cm
5 cm
3.5 cm
8.5 cm
Cursor
Targets
15 cm
25
2013‐10‐16
Latency applied
+
+
+
1 ms
10 ms
+
25 ms
50 ms
First experiment
x
3 Target widths
3 Target distances
4 Levels of Latency
8 Repetitions
45 Participants
Results
= 12,960 total trials
Does latency affect performance?
Mean time (in seconds)
How fast is fast enough?
Latency (in ms)
26
2013‐10‐16
Index of Difficulty
Mean time (in seconds)
Mean time (in seconds)
How fast is fast enough?
Index of difficulty (in bits)
Latency (in ms)
Throughput
Throughput (in bits/s)
Mean time (in seconds)
Index of Difficulty
Index of difficulty (in bits)
Latency (in ms)
Throughput (in bits/s)
Throughput
Does latency interact with target size?
Latency (in ms)
27
2013‐10‐16
Target size
Target size
+
+
3 cm
2 cm
Cursor
4 cm
5 cm
Targets
3 cm
2 cm
Cursor
4 cm
5 cm
Targets
Mean time (in seconds)
Target distance
Does latency interact with target distance?
Distance (in pixels)
Target distance
Target distance
Mean time (in seconds)
+
Cursor
+
Distance (in pixels)
28
2013‐10‐16
Target distance
How fast is fast enough?
+
+
How fast is fast enough?
Performance: ~10ms
How fast is fast enough?
Performance: ~10ms
Perception (tap): 24ms
How fast is fast enough?
Performance: ~10ms
Perception (tap): 24ms
Perception (drag): 2ms
29
2013‐10‐16
Kindle Fire
Microsoft Surface
Nexus 7
Kindle Fire
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
142ms
Nexus 7
Kindle Fire
113ms
142ms
97ms
113ms
142ms
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
96ms
97ms
113ms
142ms
90ms
96ms
97ms
113ms
142ms
Nexus 4
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
88ms
90ms
96ms
97ms
113ms
142ms
30
2013‐10‐16
Samsung Galaxy Tab
Nexus 4
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
iPad Mini
iPad 2
Samsung Galaxy Tab
Nexus 4
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
80ms
88ms
90ms
96ms
97ms
113ms
142ms
iPad 2
Samsung Galaxy Tab
Nexus 4
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
77ms
80ms
88ms
90ms
96ms
97ms
113ms
142ms
73ms
77ms
80ms
88ms
90ms
96ms
97ms
113ms
142ms
iPad 3
iPad Mini
iPad 2
Samsung Galaxy Tab
Nexus 4
iPhone 5
Nexus 10
Microsoft Surface
Nexus 7
Kindle Fire
71ms
73ms
77ms
80ms
88ms
90ms
96ms
97ms
113ms
142ms
High Performance Input
Touch & Feedback
Perception of Latency
Performance under Latency
Architecting Performance
31
2013‐10‐16
Low Latency Touch Demo
Custom
Input
Sensor
High
Speed
Display
FPGA Processing
1 ms
Low-Latency Touch Toolkit
< 0.1 ms
Touch
Sensor
< 0.1 ms
R1
R1
R1
R1
R2
R2
R2
R2
R3
R3
R3
R3
R4
R4
R4
R4
R5
R5
R5
R5
R6
R6
R6
R6
R7
R7
R7
R7
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
32
2013‐10‐16
R1
R1
R2
R2
R3
R3
R4
R4
R5
R5
R6
R6
R7
R7
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
Low-Latency Touch Toolkit
C1
C2
C3
C4
C5
C6
C7
C8
C9
C10
Low-Latency Touch Toolkit
LowLatency
Subsystem
Touch
Sensor
Frames
Touch
Processor
Touch
Sensor
Frames
Touch
Processor
High-Latency
Subsystem
Low-Latency Touch Toolkit
Low-Latency Touch Toolkit
LowLatency
Subsystem
Touch
Sensor
Frames
Touch
Processor
State Info.
High-Latency
Subsystem
LowLatency
Subsystem
Touch
Sensor
Frames
Touch
Processor
State Info.
Display
High-Latency
Subsystem
33
2013‐10‐16
Low-Latency Toolkit: Vision
Insert Steven Animatics
Low-Latency Toolkit: Vision
Insert Steven Animatics
Low-Latency Toolkit: Vision
Insert Steven Animatics
Low-Latency Toolkit: Vision
Insert Steven Animatics
Low-Latency Toolkit: Vision
Insert Steven Animatics
Low-Latency Toolkit: Vision
Insert Steven Animatics
34
2013‐10‐16
High Performance Input
Challenges: Synchronization
Touch & Feedback
Perception of Latency
Performance under Latency
Architecting Performance
Distance to
surface
Perception response
Distance to
surface
Distance to
surface
Perception response
2 ms
Time
Time
Perception response
Take away message
24 ms
We observed increases in performance, for direct-touch
pointing tasks, by reducing latency down to 10ms.
2 ms
Time
35
2013‐10‐16
Acknowledgments
Thank you!
We would like to thank Steven Sanders and
Kim Wright for their input on the document
and study.
www.cs.toronto.edu/~jotacosta
jotacosta@dgp.toronto.edu
Ricardo Jota
2013
2013
2014
36