Implementation of adaptive behaviors in a simple insect-like robot.
Mark Lanthier · 1993
Recent approaches towards designing autonomous robots have concentrated on the bottom-up approach by programming them with instinctive behaviors in which the overall behavior of the robot emerges from the interaction of the robot with the environment. This thesis describes how these instincts can be coded using simple fixed weight neural networks which reduce the need for computational speed and power. Furthermore, with these simple low level behaviors, only simple sensors are required. The resulting robot is therefore simpler, smaller and cheaper. The potential of simply constructed robots that use a minimal complement of basic sensors is investigated. A hardware and simulated version of a robotic insect is presented which have the capability of adapting to a static, initially unknown environment. The hardware version was developed to determine if such a robot could operate in the unpredictable and noisy real world. Various problems were encountered during its development that points out the importance of developing a physical device. The robot is endowed with instincts pertaining to obstacle avoidance, wandering, vacancy and edge following. Instincts from the simulated