Terrain Recognition and Environment Modeling in Legged Robots

Abstract :

In legged biological systems, while walking, the actual proprioceptive data is compared with expected values to trigger different reflexes such as stumbling or slippage correction as fast reactions on irregularities. Furthermore the proprioceptive information is used to adapt the walking behavior to the surface quality. The contribution of this paper is to present an intelligent leg of a biologically inspired robot which will be capable to classify the ground characteristics. While moving on the ground the robot makes use of its internal data such as current, angular positions of the joints, or the energy needed for the movements to classify the ground it is walking on. Additionally sensors like accelerometers and pressure sensors in the feet are used to determine the ground friction and other ground properties. In contrast to wheeled or tracked robotic systems, a legged robot has the ability to gain a wider range of information about the ground. This is due to the higher degree of freedom inherent in a walking system, resulting in an increased sensory input. The aim of the ongoing analysis will be to discern on which underlying substrate the robot is walking and how it affects the state of the robot, thus an adaptive gait behavior is made possible. The experiments presented here are done with one leg on the Leg Test-Bed in the DFKI Laboratories. The series of experiments described is part of ongoing research in the DFKI concerning environment modeling and terrain recognition in legged robots