A COSPAL Subsystem : Solving a Shape-Sorter Puzzle

摘要: To program a robot to solve simple shape-sorter puzzle is trivial. devise Cognitive System Architecture, which allows the system find out by itself how go about solution, less than The development of such an architecture one aims COSPAL project, leading new techniques in vision based Artificial Systems, allow robust systems for real dynamic environments. developed under project remain however simplified scenarios, likewise problem described present paper. key property its robustness. Since we apply association strategies local features, behaves robustly wide range distortions, as occlusion, colour and intensity changes. segmentation step applied many known from literature replaced with associations view-based hypothesis validation. hypotheses used our are on anticipated state visual percepts. This replaces explicit modeling shapes. current chosen voting verified against true obtained manipulator actions, where reinforcement learning calculation actions. These three differences classical schemes design much more generic flexible high level On technical side, channel representation information associative terms essential ingredients system. It properties locality, smoothness, non-negativity make these suitable this kind application. paper gives brief descriptions different parts have been implemented show some examples tests.