Adaptive House Lighting Control Demo 1

In this video, Mike shows how the house learns appropriate lighting behavior.  The house has been reset, so that it does not believe that the inhabitant has any preference for light over no light.  Because turning on the light costs energy, the house decides not to turn on the bathroom light when Mike first enters the bathroom.  (We show the motion detector on the wall that detects occupancy.)  Mike must manually turn the light on.  This action is treated as a punishment signal, telling the house that the lighting level it chose was inadequate.  However, the action does not specify the correct lighting level.  When Mike reenters the room, the house realizes that the level it chose the last time was inadequate, but because it does not know the correct level, it chooses a dim intensity for the light.  (The X10 controls in the house must turn the lights on to the maximum intensity, and then dim to a low level.)  Mike again punishes the system, and after a few such rounds of punishment, Mike finds the light level adequate.  The house constantly tries to minimize the energy usage, and it is only in response to Mike's complaints that it adjusts the level upward.  

Note that in this demo, we have forced the system to treat each entry as being the same "context" as the previous entry.  Ordinarily, the house's actions are context dependent, and if  the context has changed sufficiently, the system will respond to the new context not the previous one.  Context variables include the number of times a zone has become occupied in the past few minutes, as well as other movement patterns around the house.

download lighting control demo 3

Adaptive House Lighting Control Demo 2

In this video, Mike demonstrates the operation of the anticipator neural network.  When he opens the door to the bathroom, the light turns on immediately, despite the fact that the motion detector in the bathroom has not yet detected his presence.  The anticipator has learned that opening the door is a reliable predictor of motion in the room.  At the end of the video, Mike exits the room and leaves the door open.  The light once again turns on.  This behavior is also a result of the anticipator.  Because Mike generally closes the bathroom door when he exists, the mere fact that the door is open -- along with sound in the area near the bathroom -- is a strong predictor that the bathroom will become reoccupied shortly.

download lighting control demo 2

Adaptive House Lighting Control Demo 3

In this video, Mike walks around the great room and shows that the house responds to his location in the manner that it had been trained.  When he sits at the dining table, the house turns on overhead lights.  When he enters the kitchen, the house turns on lights throughout the room.  When he watches television, the house turns off the wall sconces behind the television set.

download lighting control demo 3

Adaptive House Lighting Control Demo 4

This video shows Mike walking from his bedroom into the bathroom.  The house anticipates his entry to the bathroom and turns on the lights each entry, but decides to turn on different lights the first time the bathroom is entered versus the second time.

download lighting control demo 4

The Adaptive House on HGTV's Extreme Homes

download segment