A novel algorithm for collision-free navigation of a mobile robot on the campus road with pedestrian roaming is described in this paper. Proposed method uses the RGB-D depth sensor, utilizing optical flow estimation and object detection, to predict pedestrian locations. Given these environmental uncertainties, we present a Velocity Obstacle (VO) algorithm based on mobility rules to calculate the velocity has been presented. It is proposed to use the Markov Decision Process (MDP) for decision-making (to maneuver the robot whenever it approaches the target). The proposed algorithm is a hybrid combination of deep learning and model-based techniques and provides better results in terms of navigation time and collision avoidance success rate than conventional algorithms. The real-time performance of the proposed algorithm is highlighted using a real-world dynamic scenario for an up-to-date kid ride car that has been redesigned to work as an autonomous mobile robot.

本文介绍了一种新型的校园道路行人漫游移动机器人无碰撞导航算法。所提出的方法使用 RGB-D 深度传感器，利用光流估计和目标检测来预测行人位置。鉴于这些环境的不确定性，我们提出了一种基于移动规则的速度障碍（VO）算法来计算速度。建议使用马尔可夫决策过程 (MDP) 进行决策（在机器人接近目标时进行机动）。所提出的算法是深度学习和基于模型的技术的混合组合，在导航时间和碰撞避免成功率高于传统算法。使用最新的儿童乘坐汽车的真实动态场景突出了所提出算法的实时性能，该儿童乘坐汽车已重新设计为自主移动机器人。