1 of 28

Towards Usable Attribute Scaling for Latency Compensation in Cloud-based Games

Edward Carlson, Tian Fan, Zijian Guan, Xiaokun Xu and

Mark Claypool

2 of 28

Introduction

Amazon Luna

Google Stadia

Microsoft X Cloud

Attribute

Scaling

2

Additional challenge:

Latency

3 of 28

Attribute Scaling Inspiration

Latency

3

4 of 28

Attribute Scaling Inspiration

  • With latency 🡪 game more difficult!
  • Adjust game to keep player performance “in the zone”

Too hard!

Too easy!

Just right!

Game Difficulty

Fun

Game with

no lag

Lagged game with

compensation

Lagged game

(too hard)

4

5 of 28

Attribute Scaling Example – Flappier Bird

5

6 of 28

Attribute Scaling Options

  • Precision
  • Deadline
  • Impact
  • Predictability
  • Required Action Rate

​

  • This paper:
    • Does it work? Performance with latency, scaling
    • Can we model? Generalized use by game designers

6

7 of 28

Methodology

  • Develop games with scaling action
  • Conduct user study
  • Model attribute scaling
  • Evaluate model in game
  • Develop engine API
  • Deploy
  • Evaluate

7

This paper

8 of 28

Develop Games – �Catalyst and Nova

8

9 of 28

Catalyst Scaling

9

Scale 1.0

Scale 1.5

Scale 2.0

10 of 28

Nova Scaling

10

11 of 28

User Study Parameters

11

Catalyst

Nova

12 of 28

User Study Procedure

  • Demographics survey
  • Catalyst / Nova
    • Practice
    • Play round with lag
    • Shuffle settings: latency, difficulty, scaling
    • Repeat

​

12

13 of 28

Demographics

13

14 of 28

Player Performance – �Accuracy

Nova

Catalyst

14

15 of 28

vs. Latency

15

16 of 28

vs. Scaling

16

17 of 28

vs. Difficulty

17

18 of 28

ANOVA

18

Catalyst

Nova

19 of 28

Model

19

Catalyst

Nova

R2 0.95

R2 0.69

20 of 28

Using the Model – �Scaling based on latency

20

Catalyst

Nova

21 of 28

Using the Model –� Nova

21

22 of 28

Conclusion

  • Cloud-based game streaming 🡪 new latency compensation
    • This paper: attribute scaling
  • Built games, tested parameters, derived models
  • Accuracy varies with
    • Latency, Scaling, Difficulty
  • Models
    • Catalyst – R2 0.95
    • Nova – R2 0.69

​

​

22

23 of 28

Future Work

  • Develop games with scaling action
  • Conduct user study
  • Model attribute scaling
  • Evaluate model in game
  • Develop engine API
  • Deploy
  • Evaluate

23

24 of 28

Acknowledgements

  • Catalyst
    • Alejandra Garza, Joseph Swetz, Cameron Person, Adam Desveaux and James Plante
  • Nova
    • Michael Bosik, Nina Taurich, and Alex Hunt
  • Google Stadia
    • Brian Clark, Doris Hung, Elisabeth Morant, Philip Lamoureux

24

25 of 28

Towards Usable Attribute Scaling for Latency Compensation in Cloud-based Games

Edward Carlson, Tian Fan, Zijian Guan, Xiaokun Xu and

Mark Claypool

Thank-you for your attention!

26 of 28

Extra Slides

27 of 28

Local Latency

Time

Mouse

clicked

Outcome

visible

327

…

377

27

28 of 28

QoE each round

28