Kinematic andDynamic Adaptive Control ofa Nonholonomic Mobile Robotusing aRNN

Inthis paper, anadaptive neurocontrol system withtwolevels isproposed forthemotion control ofa nonholonomic mobile robot. Inthefirst level, a recurrent network improves therobustness ofa kinematic controller andgenerates linear andangular velocities, necessary totrack areference trajectory. Inthesecond level, another network converts thedesired velocities, provided bythefirst level, into atorque control. Theadvantage ofthecontrol approach is that, noknowledge aboutthedynamic modelisrequired, and nosynaptic weight changing isneeded inpresence ofrobot's parameters variation. Thiscapability isacquired through prior 'meta-learning'. Simulation results aredemonstrated tovalidate therobustness oftheproposed approach. Index Terms-Nonholonomic mobile robots, adaptive con- trol, recurrent neural networks, meta-learning.