Research On Pose Estimation And Obstacle Avoidance Of Indoor Mobile Robots

Indoor mobile robot has a wide range of applications,which can not only improve the production efficiency,but also reduce the human cost.With a variety of input sensor information,the robot can achieve autonomous navigation in the environment with or without prior information.Lidar and camera are two commonly used sensors for autonomous navigation.Lidar can make accurate measurement,while the camera can sense the rich information in the environment.The robot needs to use this information to realize real-time pose estimation and obstacle avoidance.Pose estimation solves the problem of "where" the robot is,and real-time obstacle avoidance solves the problem that the robot reaches the target along the collision free path.In order to make the pose estimation method feedback real-time and accurate pose information on the basis of meeting the computational complexity,and realize dynamic obstacle avoidance in the environment with and without prior information,the following main research contents are proposed in this paper:In order to meet the requirements of fast and accurate pose estimation for indoor mobile robots,this paper proposes a pose estimation algorithm based on progressive scan matching.The algorithm can estimate the pose in real time in the established grid map.The algorithm generates rotation samples and progressive translation samples respectively to reduce the computational complexity and ensure the accuracy of the results.The introduction of nonlinear optimization method also further optimizes the pose estimation results to achieve more accurate results.In the process of creating raster map,the gray value of the grid is updated step by step by using the pose estimation results based on progressive scan matching.Aiming at the problem of pose estimation error caused by occlusion of important features,a visual odometer assisted pose estimation algorithm is proposed in this study,which uses the rich information of the image to compensate the pose estimation error.The algorithm needs to preprocess the input image to improve the operation efficiency and reliability,and then the pose of the robot is estimated by matching the feature points.Then the mathematical statistics method will be used to judge whether the sample score distribution based on progressive scanning matching is normal,and then to evaluate whether the pose estimation result is credible.When the results based on progressive scan matching are not credible,the maximum error direction of pose estimation should be determined according to principal component analysis,and the error compensation should be made in this direction by using the results of visual odometer to ensure the stability of tracking results.In order to solve the problem of obstacle avoidance in known environment,a model-based obstacle avoidance algorithm based on prior information is proposed in this paper.The static layer of multi information expansion map needs to be preprocessed,and other semantic layers update each time the sensor information is received.Different obstacle information is marked and updated in the corresponding semantic layer and added to the static layer.After preprocessing the current sensor information and saving it in the map,the obstacle avoidance algorithm judges the obstacle avoidance status according to the information in the multi information expansion map,and adopts the corresponding obstacle avoidance strategy.Aiming at obstacle avoidance in unknown environment,this paper proposes a Qlearning obstacle avoidance algorithm based on humanoid reasoning without prior information,which enables indoor mobile robots to avoid obstacles effectively without prior knowledge.The algorithm defines the state description and evaluation function of the robot according to the humanoid perception,so that the robot can avoid the dynamic obstacles in front of it.At the same time,the dynamic window is introduced into the behavior definition,so that the mobile robot can always run under the condition of its kinematics.This method does not need to model the environment,but also takes into account the kinematic performance of the mobile robot,which can meet the requirements of obstacle avoidance in unknown environment.In the experimental study,firstly,the construction of mobile robot and sensor calibration are briefly introduced.The accuracy and real-time performance of pose estimation is verified in the experiment based on progressive scan matching.Visual odometer aided pose estimation experiments verify that the algorithm can compensate the pose estimation error based on progressive scanning matching in occlusion environment.The performance of multi information inflation map and the effectiveness of obstacle avoidance algorithm are verified in the obstacle avoidance experiment based on known model under prior information.In the obstacle avoidance experiment based on humanoid reasoning without prior information,it is verified that the robot can avoid obstacles in dynamic environment under its kinematic constraints.The portability experiment verifies the feasibility of the proposed pose estimation and obstacle avoidance method in other robot platforms,and the universality of the method.