Localization And Navigation Of Home Service Robot In Considering Of Humam-Robot Harmony

Localization and navigation of service robot not only needs basic navigation algorithm, but also takes some mental facts like comfort and satisfaction of human in actual environment into consideration. Traditional navigation of robot focuses on autonomic movement through perception-decision mode, seeks for the accuracy of navigation and the ability of anti-jamming. It rarely involves the detection of activity pattern and the understanding of behavior intention, lacking the property of living friendly and harmony with human. In this dissertation, a series of novel methods and key technologies are explored and presented to achieve high-level capabilities of localization, navigation and robot control and improve the intelligent level of service robot in the semi-structured dynamic home environment, where a distributed sensing network in intelligent space is employed. The main contents focused in this dissertation are listed as follows.(1) Focus on human movement recognition in domestic environment, a high accuracy and robust human detection and tracking method is proposed based on the cooperation of multi-sensor. First, several distributed laser radars are employed to build the human detection platform which makes the system has the capability of multiple target tracking. In order to provide the ability of target recognition for the tracking system, color features and contour of the target are extracted and then integrated into the tracking system. Finally, this mechanism not only enhances the perceptual ability of the service robot but also solves the problems such as the localization of the serviced body and status monitoring fundamentally.(2) To achieve reliable localization and navigation of service robot in complex and dynamic domestic environment, a novel coarse-to-fine localization scheme is proposed by integrating the scene recognition with Monte Carlo method, where particles injection approaches are studied fully. In the scene recognition stage, invariant features are detected based on SURF algorithm and visual vocabulary is then built. Using visual words, all the scenarios in domestic environment could be represented. To overcome the drawback of BOW on image representation, this section proposes an effective statistics-based geometric consistency check method, which improves the accuracy of scene recognition. In the localization estimation stage, the current state of the particles is monitored using information entropy (IE) and effectiveness sample size (ESS), which establish the basis for particle injection. Robot localization results show that the proposed localization scheme could achieve fast and robust localization in the complex and dynamic environment.(3) In view of the traditional path planning methods are hardly to deal with the navigation problems in the complex domestic environment, a novel robot navigation method based on human walking trajectory is proposed. First, the description pattern of the human trajectory is discussed on the basis of the human detection, localization and tracking. A key point extraction method based on kernel density estimation is proposed. Human walking trajectories are expressed using key points, and then classified into different categories according to similarity measure on space distance, direction and velocity. Second, to avoid the aimless search of path planning in the global space and improve its efficiency, an improved RRT-Connect algorithm based on trajectory gravity function is proposed. This method realizes that search scheme tends to plan on human walking trajectories. Finally, path planning in double layers is conducted based on topology-grid representation environment model, which increases the efficiency of navigation on the basis of safety.(4) A service robot navigation and control method is proposed in this dissertation based on the human motion model analysis according to the assumption of human behavior has certain regularities in domestic environment. First, human motion modeling method is studied intensively based on the observation sequence of pedestrian movement. Second, multi-constraint path planning pattern that fused the fashion of human daily habits is then put forward. Finally, combining with detection-deliberating-reflection composite behavior patterns, service robot collision avoidance scheme is designed based on multilayer perception model. This section also implements of the corresponding motion control which makes the robot has the ability of initiative collision avoidance and increases the harmony between human and robots.