Modeling of Leg Soil Interaction using Genetic Algorithms

Abstract :

In the field of legged robotics, the use of walking and climbing robots becomes very useful for extraterrestrial applications, e.g., collection of samples from lunar crater beds. To efficiently simulate such a space mission and to enable a rational design process, a realistic robot leg–soil interaction model is required. In this paper an approach is presented that constructs such a model using a genetic algorithm that evolves an artificial neural network using experimental data. The data is collected through a series of experiments performed with an industrial robotic arm equipped with a six axes force/torque sensor and a state of the art walking and climbing robot foot. The genetic algorithm evolves the structure and the parameters of the neural network which is represented with an indirect graph. In this paper only the modeling of the contact normal forces is presented, but the approach can be easily extended to include also the lateral force. The paper describes the neural network, the genetic algorithm, and the indirect graph representation. Moreover, the integration of the model into a full rigid body robot simulator is presented.

Keywords :

Terrain interaction, legged robots, simulation, genetic algorithms, neural networks