So I've been thinking a lot about my work recently, as I've started the first implementation phase of the algorithm that I want to try out. Sadly, the more I learn about the algorithm at the moment, the more I worry about its applicability.
Let me exsplain my current thinking about robotics, security robotics and AIS.
Robotics suffers from a real identity crisis. Essentially robotics is a tool. One designs robotics to solve a specific problem. Many researchers have got themselves caught up solving problems like "SLAM" simultaneous location and mapping. This seems like a largely futile pursuit. Why? Because, quite simply, most applications in the real world, simply don't need it. For the most part, building a simple map is something that can be done once, and then ignored. Constantly updating that map is largely irrelevant. So I chose to look at a specific problem, (or as I will reveal, two specific problems.) After much deliberation I chose security, largely because it has a large scope for software improvements, rather than hardware improvements, something that I am not allowed to do as part of my computer science-based PhD.
Security robotics has two key advantages over static sensors. Firstly the static sensor can be avoided, planned around and/or obstructed. A mobile sensor can be unpredictable and manouvre around obstructions. Secondly, a security robot can be mounted with higher quality, short range sensors, that can be brought to the target being observed, short range sensors aren't much cop when static, as the area covered vs the cost is pretty poor. Security robotics is essentially two classical problems rolled into one, routing and classifying. The routing problem is one of making the best use of the robots that you have to cover the areas that you need, but at the same time, you want to prioritise bottle-necks, (entrances and exits and the like) and avoid predictability, (you don't want a thief being able to observe the robots for a day or so and spot a pattern in the way they move.) The classification problem stems from the need to automatically process the vast volume of data that a robot with the relevant sensors can produce. A single guard in a control room simply can't process all of the feeds from multiple robots. It is necessary for a robot to be capable of automatically flagging potentially dangerous situations.
This is where AIS comes in. The immune system is designed to identify threats in an environment full of confusing signals, when the threat is quite determined to avoid being detected. It seemed like a good idea to see what algorithms from AIS I can steal to perform these routing and classification tasks. But the more I learn, the more worried I become.
I will save my explanation why for another time, as this post is already quite big, and my lappy battery is running out.
Have a good one!
