Animal Robot Speech Navigation And Exquisite Controlling Method

Animal robot system by means of brain-computer interface technology, which integrated bio-logical intelligence and machine intelligence together, to some extent, has achieved a lot comple-mentary advantages, and has become one of the research highlights in the field of hybrid intelli-gence. Compared to machines, animals are good at perceiving environmental information, energy acquisition and storage, and superior flexibility of movement. But precisely because of its flexibil-ity and autonomy movement, the accurate and efficient automatic control of bio-robots becomes a big problem.In this paper, we take rat-robot as our research object, and focus on the problem of speech nav-igation and automatic controlling. For speech navigation, we use the technique of dynamic time warping to develop a precise and efficient isolated words recognition system, and successfully achieved the speech navigation task for a rat-robot. And for automatic controlling, a skeleton point motion detection method and a motion state and control instruction prediction method based on dynamic Bayesian network are presented. These methods and system fully integrated and reflect-ed the machine intelligence, animal intelligence and human intelligence. Specifically, this paper include the following aspects:1) Via the technology of isolated word recognition, we accurately and efficiently convert voice commands to control instructions, thus extends the control interfaces of animal robots. The accu-racy of isolated word recognition achieved 100% while the time consumption is less than 130ms. And we successfully achieved speech navigation on a cross-shaped maze and a sand table;2) By analyzing the morphological and behavioral characteristics of rats, skeleton extraction based state detection algorithm is proposed. This method takes into account the rat’s subtle move-ment kinematics and willingness, and fully exploits its’flexibility and autonomy features. We real-ized automatic control of the ratbot by combining this state detection method and simple artificial rules;3) Using Dynamic Bayesian Network model for motion and control modeling. This takes advantage of the timing and continuity of rats’movement, making automatic control and navigation more accurate and efficient. The accuracy of state and command prediction can achieve as high as 86.75% and 97.09% correspondingly.