We conducted a 60-second experiment (144 frames of simulation). The initial condition is a dark room and the LED input setpoint is [0.3, 0.3, 0.4] for all fixtures. External disturbance was not included in the experiment, and the controller parameters were specified with different weights on chromaticity uniformity and intensity uniformity (˛Uc = 1, ˛Ub = 0.01) and a weight on energy ˛E that is 0 from t = 0 to 30 s and ˛E = 0.04 from t = 31 to 60 s.
The top row of Fig. 18 compares the color sensor readings and LED inputs for three situations: the actual measurements from the physical space (left column), the simulated measurements in Experiment 4 with the calibrated color sensor model (middle column) and the results of Experiment 4 using the non-calibrated color sensor model based on the orthographic camera used throughout Experiments 1–3 (right column). The bottom row of Fig. 18 does the same for the normalized energy consumption.
We can see that the left and middle columns of Fig. 18 are almost identical. On the other hand, using an uncalibrated sensor model results in simulated color sensor readings that differ substantially from the actual behavior. These results illustrate the importance of calibration, and show the potential for the simulation to produce both photorealistic visualizations and correct numerical results when the material properties of the room and the response of the sensor are carefully measured and included in the model.
6. Conclusions and future work
We demonstrated an interactive framework for pre-visualizing and tuning the parameters of a lighting controller, based on a combination of photorealistic simulation and advanced control algorithm design. In our preliminary experiments, we showed how the simulation made it easy for a user to interactively refine the objective function for a controller in a series of design iterations. The simulation framework easily allows fixtures to be changed or moved, sensors added, or occupants introduced. We believe the tight coupling between advanced lighting control system design and environmental simulation will be productive for lighting designers and engineers, and can potentially minimize unanticipated or undesirable lighting behavior in built environments. While our experiments involved computer-generated animations of the lighting in a room over time from a single viewpoint, the camera could be easily moved as desired.
On the other hand, the experiments in the paper only represent a few scenarios out of the many possible real world environments and controller choices. More research is needed to make the simulation framework more generally applicable, such as a tool for automatically transforming a blueprint or CAD model into the pre-computed geometry and lightmaps required to investigate controller behavior. A bigger challenge, both technical and social, is the insertion of lighting pre-visualization tools into the typical architectural design process, especially in the early phases when the choice of lighting controller might significantly impact choices about lighting fixtures or facade elements. In future work, we plan to collaborate directly with architects and lighting designers to find ways of making the simulation framework more applicable to common practice.
Many research directions follow from this initial prototype. We are currently in the process of physically measuring light fields and transfer functions in the physical conference room under construction, to make the simulation of this space even more accurate (i.e., following on from the experiments in Section 5.4). We are also in the process of accurately characterizing and simulating new prototype sensors to be designed and deployed. We expect the lighting control simulation to inform both the development of these new sensors (e.g., necessary directional and spectral sensitivity) as well as the placement and characteristics of lighting fixtures in the constructed space.
We are also exploring several directions to make the simulation more realistic. These include moving beyond RGB sources and sensors to multispectral responses; incorporating wall-mounted sensors or sensors not collocated with fixtures; modeling occupant tracking with ceiling-mounted time-of-flight sensors, and exploring dynamic desired light fields (e.g., that adapt to moving occupants or changing weather).
A key unresolved problem, well beyond the scope of this study, is that of deciding the “right” time-varying light field for a given environment and use case. A control algorithm can be designed to accurately reach a desired setpoint, and simulation can do an excellent job of visualizing the results, but determining the setpoint itself is quite challenging. For example, the setpoint could vary according to the number, position, and pose of occupants, their activity (e.g., working in small groups, holding a discussion, watching a presentation, watching a film), and the time of day (e.g., using circadian theory to expose the occupants to warmer colors after sunset). The situation is further complicated by the challenges of designing a controller that can simultaneously satisfy the subjective preferences of multiple users, and the practical need to override the controller if the occupants find the result unsatisfactory [6]. Our continuing investigations in this area will engage lighting designers to help address these difficult questions.
Acknowledgements
This work was supported primarily by the Engineering Research
Centers Program (ERC) of the National Science Foundation under
NSF Cooperative Agreement No. EEC-0812056 and in part by New
York State under NYSTAR contract C090145. Thanks to Brandon Andow for his perspectives on architectural and lighting design, and to the anonymous reviewer for detailed suggestions on improving the presentation of the paper.
Appendix A. Supplementary data