In animal behavioral biology, there are certain cases in which an autonomous observing/training system can be useful. 1) Observation on certain species all times or for specific events, which can happen at any moment, but not often. 2) Long-period intensive training on animals to make certain experiments possible. 3) Training and testing animals without human interference to get rid of cues and biases induced by human. The goal of this study was building a system called CATOS, Computer Aided Training/Observing System, for above situations and, also, the system was tested with a case of training cats to press three buttons discriminantly depending on three different sounds(human speech).
CATOS's main advantages are as below. 1) It is based an open-source libraries, thus it is adjustable and extendable without much of difficulties in terms of accessible source code and extra cost for various different training situations. 2) It has general (visual & auditory) observational features unlike other operant conditioning chambers already introduced in Animal behavioral biology. 3) It does not have to separate animals from its group or space, while many other systems in this field to do so.
Software of CATOS uses Python as a main language, it utilizes Python's multiprocessing package to launch video-in/out, audio-in/out, and other processes to take of animal observing and training. It uses several external libraries such as numpy, scipy, pyAudio, pySerial, matplotlib, and OpenCV. The OpenCV library takes care of visual perception of the system. The auditory perception of the system was tested as a proof of concept by using HTK (Hidden-Markov-Model Toolkit, http://htk.eng.cam.ac.uk/).
Hardware of CATOS is composed of computer, touch-screen, camera, loudspeaker, microphone, custom automatic feeder, small sensors such as a photocell and a temperature sensor, a microcontroller for controlling various sensors and actuators, and the protective case.
Each component and technology of this system is not novel and I do not argue this system is innovative or novel approach, but rather it was built for practical reasons such as reducing data storage and human efforts in animal training and observation. It was tested with two domesticated cats for about six months and it successfully trained one of those two cats to press one button among three buttons on a sound stimulus playback to obtain the food reward. They did not reach the final goal (>75% correctness), but they managed to pass a couple of steps showing some learning effects, and the system acted as a trainer successfully for this period.