Research On The Development Mechanism Of Robot Based On The Dynamic Structure Network

The development mechanism of robot is similar to human brain, and it has certain cognitive learning ability to the new environment and new things. By accumulating and learning experiences, robot has similar human’s behaviors, which shows cognitive development performance of robot. It has researched the cognitive learning and development ability of the bionic model in robot’s sportive balance control. The main research achievements are as following:1) A growth algorithm of dynamic structure neural network has been put forward, with the two-wheeled robot as the experimental platform, the simulation experiments has been done. It conducts freedom of movement balance control experiment, which verifies that it has good anti-interference ability and effectiveness.2) The structure size of neural network will be too large by constant increase, a kind pruning algorithm of dynamic structure neural network is conceived for the problem, and it estimates the behavior by evaluation function, at the same time, it produces a signal of reward or punishment. The point balance control experiment has been completed, and it verifies that the effectiveness and the robustness of the improved algorithm and the performance of the robot autonomous development.3) To optimize two algorithms above, it combines the Boltzmann machine with bionic autonomous learning algorithms. It designs a kind of dynamic bionic model. Fixed-point and freedom of movement balance control experiment are conducted respectively, and it reflects the independent development ability of the robot.The bionic model of dynamic structure, which is proposed, simulates the skill formation process of the human or animal from the perspective of self-organization and self-adaption. The bionic model of dynamic structure is compared with the model of growth and pruning algorithm, and it further improves the learning speed and precision of the development of robot system.