任务驱动的发育机器人研究

Task-driven developmental robot

针对现有的发育机器人算法不能有效区分任务的缺点,提出一种任务驱动的发育机器人范式。该范式以BP神经网络为载体,将不同环境下的不同任务分开存储,每个任务中存储对应网络训练的权值与阈值,在执行任务时,会调用相应的参数重构神经网络,计算实时的输出。实验结果表明该范式不仅可以体现发育机器人的基本思想,满足机器人实时性的要求,同时能够有效地解决多任务之间冲突的问题。

Aiming at the problem of ineffective differentiate tasks of existing developmental robotics algorithms,a novel task-driven diagram is proposed.In the algorithm,BP neural network is used,and different tasks in different environments are not saved together.Weights and thresholds of neural network are saved in corresponding task.The parameters will be used to reconstruct neural network for computing outputs when execution of the corresponding task.The experiments results show that the diagram not only can embody the idea of developmental robotics and meet the acquirement of real time,but also can effectively avoid conflict of different tasks.