blackrimglasses.com

As I posted earlier, Amy and I are going to be moving to a new house this summer in Woodland Hills. We’re moving from a multistory (four in fact) townhouse that we gutted and renovated into a sprawling ranch style house that was recently renovated, albeit not specifically to our taste.

The work we’ll do to the new home in terms of renovation is a subject for Amy’s blog (a hint, it’ll be just some floors and paint right now), but what I’m focused on is the next phase of my DIY Smarthome work.

With the Burbank house, the Smarthome aspects were focused on command/control architectures, all human integrated. Meaning: the lights responded to explicit input, as did any other collective control of the house systems. We had only integrated lighting besides. It lacked any comprehension of the heuristics of our every day living.

With the new home, my goal is to bring a sense of heuristics to the Smarthoe approach. To do so, I need to make the system learn, and respond to set inputs.

Here are the inputs that I’m considering putting in:

iPhone Location

Using some URL hacking of MobileMe to let the system know where Amy and I are in the world, whether I’m at work, home or elsewhere. Amy wisely pointed out that she might leave her phone at home. I view this edge case just that, an edge case and not invalidating the entire precept.

Motion Sensors

The key is to use motion sensors not as a mainline input to any condition, but as a system to help us determine certain conditions.

LED light sensors

These exist for Insteon systems, and can determine if a light is turned on or off. A good example of usage for this is the stereo receivers light, or the DirectTV boxes.

Audio Sensor

There is another sensor that lets you feed in line-level audio and trigger an Insteon signal based on that. You could, theoretically, chain this off a stereo aux output to determine if there is activity in the living room.

Garage Door Sensor

Will sense state changes (open/close).

Door Sensors

Likewise, except on key interior and exterior doors.

Twitter

Just like MobileMe URL hacking, using our latest tweets and their geotags to determine where we are.

There are a few others I’m sure, but the goal here is to provide a set of inputs that together can be analyzed to answer two questions: who’s home and where are they.

The idea I have is to feed all these inputs periodically into a back-propagating neural network which is trained to give certain outputs given states of inputs. For instance, we could train it that if I’m home, Amy is not, and the doors are closed and I’m watching TV, the house should turn off all lights but the living room and adjust the thermostat to my liking.

Likewise, if it detects no one is home, the temperature outside is above 90 and the doors are all closed, it’ll keep the thermostat adjusted for the dogs.

Indigo, from Perceptive Automation is apple-scriptable, and I’m not sure if anyone has made a neural-net in AppleScript. However, I have written one in Python so I’ll most likely feed the variables from Indigo out, and periodically (on Cron?) feed its outputs back into Indigo to trigger certain action-groups, which are like macro’s for Insteon and other systems.

Stay tuned for more articles on my DIY Smarthome Project!