Classification and Identification of Environment Through Dynamic Coupling

This paper presents a methodology enabling robotic systems to classify and identify their environment according to the mechanical properties of the local contact dynamics. Described approach employs existing proprioceptive sensors and requires no additional specialized hardware. Identification process is performed in real-time with temporal resolution of measurement updates determined by the periodicity of the limit behavior. While the basic concept has a wide application spectrum, our discussion focuses on terrestrial locomotion where contact properties, such a compliance, damping, sheer friction and surface topology, are important environmental markers. Accurate identification of environmental parameters enables two types of applications. In behavioral control, availability of measurements on environmental parameterization can facilitate better adaptation of actuation strategy. In localization and map building applications, such mechanical characteristics of the environment, which are typically hard to attain, can serve as a new set of classifiers. Presented approach is founded on the observation that locomotive behaviors, and particularly the dynamic ones, emerge from the interaction between the active actuation actions of the mechanism with its environment. To evaluate our concept in a systematic fashion we constructed a simplified numerical model of a dynamic hexapod robot. We present results on numerical simulations and outline a path for a physical implementation on dynamic hexapod robot.