The original BEAM movement took place from 1990 until 2000. It seems that the principle has come to a dead end. Nevertheless, BEAM robots are looking interesting and it makes sense to take a closer look into the self understanding of the 1990s DIY robot movement.
At first it should be mentioned that in the 1990s there was the last great AI winter available. It was the time before the advent of deep learning and most robotics projects were canceled. During the 1990s it was impossible to build biped robots anad there was general pessism available if AI can be realized at all. BEAM robotics can be seen as a crisis phenomena towards the AI winter in the 1990s. The main principle is, to build robots without a brain. They are equipped with actuator and a battery but they do not have a program and they can't decide something. Instead the control is realized with random oscillators known as nv neurons.
For reason of simplication it makes sense to assume, that a BEAM nervous network is a normal 555 tiimer IC which generates a sine osscilator signal which is send to the servo motors. This signal will produce the forward movement. Certain input signals like touch signals and light signal can modulate the signal.
From a technical standpoint BEAM Robotics doesn't belong to AI history but its the opposite. AI is about a computer program which makes a computer think like a human while BEAM Robotics is an electronics project without such a software.
Because the mentioned 555 timer chip can be emulated in software very well, there is no need to buid robots in physical hardware Instead the brain of a robot can be realized as a single line of code in python which is an osscilator. The same movement of a beam biped walker can be emulated with a modern arduino microcontroller and this might explain why BEAM robotics has disapeared after the year 2000. Nevertheless, a classical analag BEAM robot looks more interesting. In the 1990s most examples were not realized with the 555 IC but with analog circuits with dedicated resistors and capicitors which produces a chaotic behavior and a chaotic layout both.
Nevertheless it should be mentioned that even physical BEAM robots from the 1990 were never intelligent or were able to do useful tasks. The movement was either realized with remote control or with randomized osscilator generators in hardware. Modern robotics can learn a lot from this approach. BEAM robotics is referenced as a minimalist attempt in creating robots. It was an ultra low cost technology without any complex hardware nor dedicated theories. Most robots were nothing but a paperclip mounted on a servo motor. So its some sort of physical artificial life.
After the year 2000, beam robotics has felt out of fashion in favor of computing oriented AI disciplines like Deep learning, Semantic web, and microcontroller based robotics. Especially the last one has replaced former DIY BEAM robotics entirely. Today entry level robots are always computer controlledd and they are programmed to fulfill simple tasks like line following or maze solving. Such a microcontroller controlled robot is more flexible than former beam robots. It allows to explore different approaches including chaotic pseudo random behavior and its much easier to rewrite the hardware. The main problem with analog CPU free robotics is that every circuits requires new parts and these parts are connected with solder similar to amateur radios in the 1950s. This is less interesting than building robots with a microcontroller and realize the logic in software.
But from an aestihic perspective, modern arduino based robots are looking a bit boring. Most examples are realized with a brick on top of a wheel chasis which is standardized. Another major criticism against microcontrollers is that this technology forces the programmer to think into a certain direction. The typical question is mostly which sort of java program is needed to fulfill a task. This perspective prevents that possible non programming alternatives are investigated.