script

Presentation:  Neural Networks in robotics

Shortform:

Realistic mobile robots demand navigation abilities to effectively plan efficient routes and reach target positions in unstructured and dynamically changing environments. The fundamental ability is position estimation using data gathered by interacting with the world. Such environments and their perceptual signatures (sensor readings) which associate the current robot's position , are hardly to be known a priori at the design time of the robot. Therefore, It is mandatory for these robots to incorporate learning capabilities so as to perform in a way that excludes human operators.

The self-organising map (SOM) is an efficient algorithm that can build 2D grid representations of similar perceptual signatures which are close to each other in the input space, onto contiguous locations in the output space. The most interesting aspect of SOMs is that the process takes place without supervision.