Robots are very expensive and still not very versatile. It would be simpler to install control circuits with remote controls for those things in a house that can be turned off and on.

As a rule, I disapprove of ideas geared at shifting responsibility over to technology. Learning good habits would eliminate a multitude of otherwise useless ideas.

Like [sand], though, I believe controls at each of these interfaces would be a lot easier to implement, and less costly. Also, I would just make it a matter of sending a single message to handle everything, rather than separate messages for each contingency.

A cost effective means of doing what you describe loosely exists. See http://www.x10.com/homepage.htm for a means to interface your SMS device with the devices in your home. ---- Here are some of the reasons why a human-like robot is currently difficult to implement. To simply identify one object from another at different angles of observation requires experience. Experience is a function of memory and interpolation. Since from any point of view, the number of different valves known is in the hundreds of thousands. Yet from a human perspective, humans reason out what a valve is. So experimentally, we can manipulate most valves with a little effort. So far, with machine vision technologies, we have to teach a machine intelligence program each, AND EVERY different type and style of valve. If a valve is said to be only turning off, or turning on, and their are only 1000 valves and 1000 environmental variables for each to be considered (only a tiny percentage of what actually exists), and there are thousands of lighting and environmental inconsistencies, that would be 2 raised to the power of 1000 as the possible choices for comparing against. The current view would need to be compared to every valve picture in in every environmental variable in the database and determine if it exists or not. That is over 1 followed by 300 zeros as possible solutions. It would take a computer running at 5 gigahertz and assuming one evaluation per clock cycle, over 6.8 followed by 277 million years to come up with a result. In practical terms, you just can't compare what you see to a database and come up with a result. The system must include deductive capabilities. There are many persons working on such problems. Any person here could happen upon a viable solution for such reasoning. See

http://wtec.org/ConvergingTechnologies/Report/NBIC_report.pdf#search=%22%22university%20of%20utah%22%20intelligence%20-koza%20evolutionary%20genetic%20-paxton%20-gribbin%20-strangg%20-moran%20-padgett%20%22artificial%20brain%22%22