移动机器人室内场景主动识别的强化学习方法

A reinforcement learning method for autonomous recognition of mobile robot in indoor scene

为了在传统场景分类器基础上进一步提高场景识别准确率,提出了一种采用Q学习(Q-learning)实现室内场景主动识别的算法.该算法采用极限学习机(Extreme Learning Machine,ELM)与反向传播梯度下降相结合的方式近似Q-learning值函数的神经网络.算法基于Q-learning动态地学习场景识别率最高的机器人朝向角,使机器人能够自主获取多次更为可靠的传感器信息并将对应识别结果融合,进而提高场景识别准确率.将算法应用在移动机器人场景识别中进行实验,结果表明该算法可以有效提高场景识别准确率.

In order to further improve the accuracy of scene recognition on the basis of the traditional scene classifier,an algorithm for automatic recognition of indoor scene using Q-learning is proposed.The algorithm fits the neural network of the Q-learning value function by the way of combining Extreme Learning Machine(ELM)and back propagation gradient descent.The algorithm can learn the robot orientation with the highest scene recognition rate dynamically based on Q-learning,which makes the robot acquire reliable sensor information several times autonomously and fuse corresponding recognition results,thus improving the accuracy of scene recognition.The algorithm is applied to the experiment of mobile robot scene recognition,and the result shows that this algorithm can effectively improve the accuracy of scene recognition.