Importance of Real-Time: The Effect of Latency in Scientific Visualization Nathaniel Jones | Sustainable Design Lab | Massachusetts Institute of Technology | New England Symposium on Graphics | April 23, 2017
Problem
Experiment
Glare in buildings is difficult to predict because photometrically accurate simulations are time consuming.
Forty participants completed a design exercise to find and mitigate glare issues in two office spaces.
Architects are less likely to make use of building performance simulation if it is not interactive.
They used two tools:
False color display reveals the dynamic range of interior luminance distributions. 103 102 cd/m2
Time & date (sun position) Camera view Shading devices
They used physically based rendering to identify glare.
Glare caused by direct and reflected sunlight reduces productivity and leads to increased energy use.
10
They could interact with the space by changing three factors:
AcceleradRT
DIVA-for-Rhino
Progressive path tracer
Radiance-based ray tracing
GPU-based
20-30 seconds per rendering
5 fps
Industry standard tool for architectural lighting analysis
Responds to mouse navigation
104
Results path tracing produces accurate Daylight 1) Progressive Glare Probability readings within 10 frames (2 seconds). Daylight Glare Probability (DGP) is the likelihood an individual will experience glare in a given view.
2)
Participants who saw progressive rendering interacted more frequently with the design tool. Participants could interact with the model in three ways: changing the time, shading devices, or camera position.
Source luminance Source solid angle
Guth position index Vertical eye illuminance
DGP can be predicted using global illumination algorithms such as progressive path tracing. 100%
11 AM
Results converge on measured values after many frames …
100%
AcceleradRT DIVA-for-Rhino
80
AcceleradRT DIVA-for-Rhino 75%
60
States
Mean Number of Interactions
100
40
Typical DIVA-for-Rhino simulation time 50%
25%
20 0
With real-time results, participants make decisions faster, even neglecting the time taken to run simulations.
0% Time and Date
Shading
View
0
10
Daylight Glare Probability
75%
1 PM 10 AM
50%
50
Overcast
… but early results are reasonably accurate. 0%
1
10
100
1000
10000
Frame
10
103 102 cd/m2
4)
Participants who saw progressive rendering created more optimal design solutions. Participants had two objectives:
With real-time feedback, participants proposed more designs on or near the Pareto frontier.
• Provide sufficient natural light • Reduce risk of glare More light
30%
AcceleradRT on Pareto front, 32%
Number of participants choosing this design
20%
DIVA-for-Rhino
In post-exercise surveys, participants were …
… when using AcceleradRT.
… when using DIVA-for-Rhino.
More confident in glare assessment
58%
13%
More confident in final design
58%
23%
Found task more enjoyable
63%
15%
More relaxed
63%
18%
Learned more
58%
10%
More familiar with tool
18%
55%
Preferred overall
90%
10%
Accel: >3 DIVA: 1 DIVA: 2
15%
DIVA: 3 DIVA: >3
AcceleradRT near Pareto front, 45%
DIVA-for-Rhino
10% DIVA far, 36%
5%
0%
AcceleradRT
Participants who saw progressive rendering were more confident and satisfied with the design exercise.
Accel: 3
More glare
Daylight Glare Probability
AcceleradRT far, 24%
Accel: 1 Accel: 2
DIVA-for-Rhino
AcceleradRT
Not Chosen 25%
104
AcceleradRT
0%
20%
40% 60% 80% Spatial Daylight Autonomy
100%
DIVA on Pareto front, 14%
DIVA near Pareto front, 50%
Conclusions predictions of glare can be produced at interactive speeds using 1) Accurate progressive rendering, even though high-quality visualizations take longer. 2)
Real-time results change user behavior and promote interactive user response, resulting in more exploration of the problem and of design solutions.
3)
Users with access to real-time rendering tend to produce more optimal designs. At present, they outperform optimization algorithms.
4)
Users also prefer the experience of real-time feedback and feel more confident and relaxed, even when the tools are less familiar.
http://mit.edu/sustainabledesignlab/projects/Accelerad/
About Dr. Nathaniel Jones is the developer of Accelerad, a suite of GPU-based lighting and daylighting simulation tools. Accelerad has registered users in fields from architecture and sustainability consulting to psychophysics and spacecraft design. He recently completed his PhD in building technology at the Massachusetts Institute of Technology, where he researched the role that simulation speed plays in how designers choose to use tools. He chairs the IBPSA-USA Emerging Simulation Technology Subcommittee, which seeks to create awareness and promote use of emerging building simulation tools and technologies within the research, industry, and government communities, and to facilitate interaction between these communities.
[email protected]
60
In aggregate, participants analyzed more of the space when they had realtime feedback.
2 PM 25%
3)
With real-time feedback, participants moved the view to cover a greater portion of the space.
12 PM
20 30 40 Time Spent in State (seconds)
The Tesla K40 accelerators used for this research were donated by the NVIDIA Corporation.
