Hybrid architectures for intelligent robotic systems

Hybrid architectures, based on combinations of analogic, symbolic, and neural methods, are well suited for real-time applications in advanced robotics. Real-time industrial applications are mainly based on the correction of preplanned programs. So far, the planning and control modules of these kind of applications are often unable to react and/or classify un-expected events. The approach described attempts to integrate the sensor-based analogic method and the neural method into a multiple-level architecture that operates on an analogic world model, so that the action planning can be performed in a smart, reactive way. Given the task, the system builds the world model of the scenario. The reasoning and planning modules act both at the strategic as well as reactive levels, and the activated sensor-based motor strategies handle the sensorial data inputs and drive the robot controller module in the execution of the stream of motor commands. The interaction between the different levels is mainly based on the idea of maintaining and updating in real-time the world model, so that each module can locally operate on specific parts of the whole world model.