改进的Q学习算法在轨迹规划中的应用

为解决 Q 学习算法易陷入局部最优解问题, 改进了传统贪婪策略, 提出了一种分段渐近搜索策略。该策略通过动态调整策略参数, 使 Q 学习算法在学习过程中实现探索鄄学习鄄利用 3 个阶段的渐近跳转。 同时将该搜索策略应用于 Q 学习算法中, 使改进的 Q 学习算法能更快速地逼近全局最优解。 将改进算法应用于机械臂轨迹规划中, 其仿真结果表明, 该算法能稳定地引导机械臂沿最优轨迹快速到达目标位置。

Improved Algorithm of Q-Learning for Trajectory Planning

Aiming at the local optimal solution for Q learning algorithm, a segment incremental search strategy was proposed base on greedy strategy. The improved Q learning jump gradually between three situations such as explore, learn and utilize by adjusting parameters of segment incremental search strategy, and it could approach the global optimal rapidly than the traditional one when the new search strategy is applied to the Q learning. The simulation results show that the manipulator reaches the target position accurately and quickly guided by the improve Q learning algorithm.