1. Bokyung Lee, Taeil Jin, Sung-Hee Lee, Daniel Saakes
Department of Industrial Design, KAIST, Korea
Virtual Humans with Agency for Design Tools
SmartManikin:
3. It is hard to estimate comfort or use-scenarios
within the design tools.
01. Introduction
4. Human Bodies in Design Tools
02. Related Work
Designing on the human body
Mirror Mirror (CHI’16)
Tactum (CHI’15)
Human Bodies as input
BodyMeter (nordiCHI’16)
Tactum (CHI’15)BodyAvatar (UIST’13)
5. DressUp (TEI ’12)
Riding Posture (Applied Ergonomics ’17)
Ergonomic Simulations
SketchChair (TEI’ 11)
Tactum (CHI’15)Jack
Virtual Humans in Design Tools
Designing on virtual manikins
02. Related Work
6. Manipulate it directly Show it what to do Tell it what to do
How to Interact with Virtual Humans?(Badler, 1995)
02. Related Work
7. Manipulate it directly Show it what to do Tell it what to do
How to Interact with Virtual Humans?(Badler, 1995)
02. Related Work
8. Manipulate it directly Show it what to do Tell it what to do
How to Interact with Virtual Humans?(Badler, 1995)
02. Related Work
9. Make it respond
Virtual Human that responds to design changes.
Manipulate it directly Show it what to do Tell it what to do
02. Related Work
13. Four Categories of Poses
comfortable pose temporary pose possible pose
(reachable pose)
healthy pose
100°
03. Formative Study
14. Design Strategies for SmartManikin Agency
comfortable pose temporary pose possible pose
(reachable pose)
healthy pose
100°
• Simulate multiple actions to inform real-life uses.
• Respond to real-time design contexts.
• Generate poses based on comfort, except reaching.
• Simulate comfort and ergonomic separately.
03. Formative Study
15. Design Strategies for SmartManikin Agency
comfortable pose temporary pose possible pose
(reachable pose)
healthy pose
100°
• Simulate multiple actions to inform real-life uses.
• Respond to real-time design contexts.
• Generate poses based on comfort, except reaching.
• Simulate comfort and ergonomic separately.
03. Formative Study
16. Design Strategies for SmartManikin Agency
raise a tabletop
raise a shelf
comfortable
pose
possible
pose
comfortable
pose
possible
pose
vs.(typing)
(reaching) vs.
• Simulate multiple actions to inform real-life uses.
• Respond to real-time design contexts.
• Generate poses based on comfort, except reaching.
• Simulate comfort and ergonomic separately.
03. Formative Study
17. Design Strategies for SmartManikin Agency
raise a tabletop
comfortable
pose
possible
pose
vs.(typing)
• Simulate multiple actions to inform real-life uses.
• Respond to real-time design contexts.
• Generate poses based on comfort, except reaching.
• Simulate comfort and ergonomic separately.
03. Formative Study
19. Pose Capturing Setup
Kinect A Kinect B
Two Microsoft Kinects were installed to capture 3D human poses.
Height
Changing
Desks
04. Pose Capture Study
20. Conditions for Comfortable Pose Capturing
Sitting TypeAction Type Item Heights Item Positions
1. Typing with laptop
• Two Hands motion
2. Watching Monitor
• Hands free
3. Using Mouse (surfing)
• One hand action
4. Reading paper
• No computer
04. Pose Capture Study
21. Conditions for Comfortable Pose Capturing
Sitting TypeAction Type Item Heights Item Positions
04. Pose Capture Study
22. Conditions for Comfortable Pose Capturing
Sitting TypeAction Type Item Heights Item Positions
04. Pose Capture Study
23. Conditions for Comfortable Pose Capturing
Sitting TypeAction Type Item Heights Item Positions
04. Pose Capture Study
24. 3D Human Pose Capturing with Skeletons
All frames of the skeletal pose were saved in bvh file.
04. Pose Capture Study
25. All frames of the skeletal pose were saved in bvh file.
3D Human Pose Capturing with Skeletons
04. Pose Capture Study
26. Pose Feature Regression
offline process
online process
pose of
Smart
Manikin
predicted features
Pose
Generator
desk
configurations
user
information
3D Pose
Dataset
Training
Regression Functions
Identifying
Key Features
Regression
Functions
base
position
foot
position
trunk
angle
(goes higher)Body backwards
Participant 6 Participant 4
base
position
04. Pose Capture Study
27. Pose Feature Regression
offline process
online process
pose of
Smart
Manikin
predicted features
Pose
Generator
desk
configurations
user
information
3D Pose
Dataset
Training
Regression Functions
Identifying
Key Features
Regression
Functions
base
position
foot
position
trunk
angle
foot
position
Participant 1
04. Pose Capture Study
28. Pose Feature Regression
offline process
online process
pose of
Smart
Manikin
predicted features
Pose
Generator
desk
configurations
user
information
3D Pose
Dataset
Training
Regression Functions
Identifying
Key Features
Regression
Functions
base
position
foot
position
trunk
angle
trunk
angle
Participant 9
04. Pose Capture Study
29. Pose Training
04. Pose Capture Study
Interacting with
distant object
Interacting with
near object
Sitting Type
Selected
Sitting
Type
Trunk
Angles
Base
Position
Foot
Position
Regressors
Trunk
Angles
Base
Position
Foot
Position
Estimated Feature
30. 04. Pose Capture Study
Input : Desk Configuration(desk height & human height)
Output: each identified features.
• Trunk Angle
• Base Position
• Foot Position
, }
𝑝𝑏𝑎𝑠𝑒 ∈ ℜ3
𝑝 𝑓𝑒𝑒𝑡 ∈ ℜ3
× ℜ3 We should change figure
(from data scene)
𝑇𝑐h𝑎𝑖𝑟
𝑇𝑏𝑎𝑠𝑒
Train data for regressor
Pose Training
31. 04. Pose Capture Study
• Regressor Model
Fully connected multilayer perceptron
• Activation Function
Sigmoid functions
• Optimization Method
batch L-BFS optimization method
• Train & Running Time
30 minutes to train all 12 regressors
(3 regressors for 4 sitting types)
Average running time is extremely fast
(around 0.2 milliseconds)
• Regression Error
Average Size of Training Datasets: 26957
Average Size of Test Datasets: 11552
Average of Regressors: 2.1%
Regress
or output
Regressor
Input
1 Hidden layer of five cells
Pose Training
34. User Interface for SmartManikin
items
item
repository
configurable
desk
action
commandsergonomic
guidelines
Using Backrest
SmartManikin
05. SmartManikin
35. 3D Desk Configuration Tool with real-time Use-Simulation with Virtual Humans
User Interface for SmartManikin
05. SmartManikin
36. [1] Poses Respond to Desk Configurations.
Agency Design of SmartManikin
lower
body
pose
upper
body
pose
FK
IK
furniture
configuration
mannequin
height
IK : Inverse Kinematics
FK: Forward Kinematics
base
position
foot
position
trunk
angle
regressors
05. SmartManikin
37. [1] Poses Respond to Desk Configurations.
Agency Design of SmartManikin
05. SmartManikin
38. [1] Poses Respond to Desk Configurations.
Agency Design of SmartManikin
05. SmartManikin
39. [2] Poses Respond to Action Input & Item Arrangements
Agency Design of SmartManikin
lower
body
pose
upper
body
pose
FK
IK
furniture
configuration
mannequin
height
IK : Inverse Kinematics
FK: Forward Kinematics
base
position
foot
position
trunk
angle
regressorshead
poses
arm
poses
action
command
object
position
IK
IK
lower
body
pose
upper
body
pose
FK
IK
base
position
foot
position
trunk
angle
furniture
configuration
mannequin
height
regressors
05. SmartManikin
41. [3] Poses Respond to Sitting Preference / Habits
Agency Design of SmartManikin
a bstanding straight leaning
c dusing backrest not using backrest
05. SmartManikin
46. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of third-person perspective.
“It is cute! It’s like my virtual assistant!”
06. Expert Feedback
47. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of third-person perspective.
“It is cute! It’s like my virtual assistant!”
06. Expert Feedback
48. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of third-person perspective.
“It is cute! It’s like my virtual assistant!”
06. Expert Feedback
49. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of third-person perspective.
“It is cute! It’s like my virtual assistant!”
06. Expert Feedback
50. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of third-person perspective.
06. Expert Feedback
51. Expert Feedback
• Emotionally attached to the virtual manikin.
• Appreciate pose simulation generated from the field observation.
• Valuable for early stage of design process.
• Valuable for real-time quick evaluation process.
• Limitation of the third-person perspective.
eye level
06. Expert Feedback
52. Limitations
07. Limitations & Future Work
Based on
Skeleton Data
Poses Collected
during Short Period
Limited
Design Freedom
(body physique x) (long-term comfort x) (only desk scenario)
57. Bokyung Lee, Taeil Jin, Sung-Hee Lee, Daniel Saakes
Department of Industrial Design, KAIST, Korea
Virtual Humans with Agency for Design Tools
SmartManikin: