Learning Intelligent Controllers for Path-Following Skills on Snake-Like Robots

Multi-link wheeled robots provide interesting opportunities within many areas such as inspection and maintenance of pipes or vents. A key functionality in order to perform such operations is that the robot can follow a predefined path fast and accurately. In this paper we present an algorithm to learn the path-following behavior for a set of motion primitives. These primitives could then be used by a planner in order to construct longer paths. The algorithm is divided into two steps: an example-based stage for controller learning, and a controller tuning stage, based on an objective function and simulations of the path-following process. The path-following controllers have been tested with a simulator of a multi-link robot in several complex paths, showing an excellent performance.