Active exploration of a robotic workcell using contact and non-contact sensors
暂无分享,去创建一个
Robotic manipulatory systems are being developed with an ever increasing vigor for a variety of applications in hazardous or unstructured environments. Their successful utilization relies heavily on the use of sensory feedback mechanisms to guide their motions to perform a variety of tasks. The sensory feedback mechanism typically includes one or more sensors, such as vision, range, proximity, tactile, and force/torque. This research focuses on developing a robotic exploration system with emphasis on using both contact and non-contact sensors. Particularly, we intend to emphasize the distinctive qualities of haptic exploration that need to be exploited to perform a variety of robotic manipulatory tasks. When the robotic work environment is not conducive to acquiring vision, and range measurements, or tasks requiring grasping of an object, haptic exploration holds the potential for providing the necessary feedback to the robotic system to accomplish its tasks. This research presents a computational framework for an active exploration system that is characterized by a modular architecture, closed-loop control, incremental data integration scheme, flexibility, and graceful degradation. The exploration system involves acquisition, processing, and integration of sensory inputs acquired from vision, point laser range, tactile and force/torque sensors. For the purpose of active exploration, no a priori information associated with the composition of the workcell (e.g., information available from object or world models) is utilized. The exploration process results in the building of three-dimensional Half-space models of various polyhedral (convex and concave) objects encountered in the workspace. A set of sensor specific exploratory mechanisms has been designed to allow systematic and repeated used of sensory modalities for the acquisition of various descriptors. Subsequent to the process of building object models using active exploration mechanisms, the object models allow extraction of essential information required for performing tasks such as object identification, grasping, and manipulation.
The experiments include convex, concave, thin planar rigid objects, and also compliant or non-rigid objects. The experiments bring out the advantages associated with having modular architecture, closed-loop control mechanism, incremental data integration, flexibility, and graceful degradation. The quantitative analysis of the experimental results show that the 3-D models are similar to the original object whenever it is possible to easily access the individual surfaces of the object. For thin planar objects, a composite tactile imprint is generated, instead of a 3-D half space model. The experiments also demonstrate the unique features of flexibility and graceful degradation.