内在动机轮式倒立摆反应式认知系统

摘要

Aiming at the problem of poor adaptability and robustness of the conventional self-balancing control method for linear motions of the wheeled inverted pendulum (WIP ), the wheeled inverted pendulum reactive cognitive system (WIP-RCS)was established,which can produce independently self-balancing control rules of the WIP through the interaction with environment.The WIP-RCS consisted of a perception module (PM),an execution module (EM)and a cognitive module (CM).PM and EM were responsible for the input and output of the systen.CM mainly involved knowledge model and learning strategy.The knowledge model was composed of a team of continuous-action learning automatic unit and served to describe the control rules. The learning strategy was a learning algorithm motivated by uncertainty motivation and served to optimize the knowledge model.The structure,working principle and learning algorithm of WIP-RCS were described in detail.The convergence of the learning algorithm was proved in theory,and the self-learning ability of WIP-RCS was verified by simulation experiments.The adaptability and robustness of the system were discussed with the combination convenfional PID and LQR. The simulation results show that the system can produce self balancing control rules,together with good learning cognitive skills,and has better adaptability and robustness.%针对直线运动的轮式倒立摆常规自平衡控制方法存在自适应性差和鲁棒性差的问题,建立轮式倒立摆反应式认知系统,在与环境交互过程中涌现出自平衡控制规则.该系统由感知模块、执行模块和认知模块组成,前两者分别负责系统的输入与输出.认知模块涉及到知识模型与学习策略:前者由连续动作学习自动机组构成,用于描述控制规则;后者负责优化知识模型,采用不确定性动机驱动的学习算法.详细描述该系统的结构、原理和算法,理论证明学习算法的收敛性,并通过仿真实验验证系统的自学习能力.结合常规PID和 LQR算法通过仿真实验验证自适应性和鲁棒性.实验结果表明,该系统能够自主涌现出自平衡控制规则,表现出良好的自主学习认知技能,具有较好的自适应性与鲁棒性.