| When robots leave from factories to human families as our pet, playmate and other service roles to be in company with us, they will face the world that we mankind face, so they are supposed to acquire similar humanlike spatial behavior abilities. Although robot researchers made a lot of research works, robots navigation technology and social skills have got great progress, yet what robots can do right now are far from people’s expectations. Human beings and animals are able to perform perfect navigation capabilities even if they are in complex and constantly changing large-scale space and the conditions where the information they perceived is inaccurate. This motivates robot researchers to pay close attention to biological navigation mechanisms that can be implemented on an autonomous mobile robot, so as to improve robots’navigation ability.From the perspective of creatures’spatial behaviors, the difficulties in navigation system design mainly come from three global issues:dynamic user tasks and goals, varying spatial memories on the basis of personal experience, and infinitely complex space. To cope with these difficulties, how to design a simple commonly-applicable spatial representation model and a wayfinding strategy with low consumption and high efficiency is one of the challenges robot researchers have to confront with. The remarkable humans’navigation mechanism enable humans to successfully deal with these challenges. In the future, intelligent robots which will live with us should also overcome the three difficulties. Therefore, inspired by the human ability to navigate, this paper attempts to roll spatial internal representation model and the navigation mechanisms of the human into one, so as to construct an autonomous navigation system applied to robot, in order to make robot acquire humanlike navigation capability, and provide references for the solving of the above three navigation difficulties.Contributions of this thesis can be summarized in several points illustrated below:1) On the basis of Growing Neural Gas(GNG) algorithm, an topological network extraction method of the environment feasible regions is proposed. Using the growth characteristics of GNG algorithm, this method conducts feasible region information extraction from the surrounding environments by adding new node of topology network and builds an environmental map which robot can understand easily. This method is also suitable for dynamic environments due to the impacts of dynamic obstacles can be filtered and constructs a consistent topological map. This method is self-learning and adaptive. The simulation and physical experiments verify the feasibility and effectiveness of the proposed method.2) From the perspective of local motion planning method, fast and reliable local obstacle avoidance method is proposed. Distance and direction are two important evaluations for robot to judge the risk of obstacles. A 2D bubble artificial potential field (B-APF) with low computation was designed and realized. Thus, by using this field, the equilibrium problem between distance and direction was described in the numerical way and a unified risk value of obstacles was produced. Robot uses the values for rapid assessment of the risk of obstacles, and then optimizes the avoidance behavior. This method has strong ability to accommodate environment and has characteristics of low computation complexity, excellent real-time ability. Simulation results show the feasibility and effectiveness of the proposed method.3) Humans make use of the regionalized spatial knowledge in the process of navigation. Inspired by this, this thesis puts forward a region-based spatial knowledge model (RSK-Model). In the model small scale regions are grouped together to form the bigger regions at the next hierarchy level which leads to a hierarchical spatial representation structure. It is difficult for the current spatial knowledge representation model based on general graph theory to reflect the complex composition and the implying organizational structure. To cope with this challenge, then, this thesis introduces the concept of hybrid hypergraph and puts forward a spatial knowledge representation model based on hybrid hypergraph to represent the spatial knowledge network with large scales, complicated connections and nested characteristics. Presentation of one simple environment is taken as an example to illustrate the application of the spatial knowledge model based on hybrid hypergraph, hence providing a new tool and approach to the expression, organization and analysis of complex spatial knowledge representation.4) From the perspective of global wayfing strategy, human has used the regionalized spatial knowledge and adopted the ’fine-to-coarse’ way-finding strategy in the process of navigation. Inspired by this, based on the regionalized spatial knowledge model, a kind of online route planning algorithm FTC-A*(fine-to-coarse A*) is proposed, which can take different planning strategies according to the distances of environmental information. In the area where the robot is, a fine route planning will be conducted while for the distant space a coarse planning will be done. Taking advantage of regionalization feature of environment description, such a strategy can shrink the research space, thus remarkably reducing planning time and memory loading as well as lowering the motion response lags of robot. The algorithm FTC-A* can be applied to such an occasion where there is a huge scale of environment or target point frequently changes. Through the simulation experiment on MobileSim platform and the contrastive analysis of algorithms A* and HA*, it is verified that the proposed method is feasible and effective.5) On the basis of qualitative spatial representation and human navigation mechanism, a bio-inspired robot navigation system is studied. First, qualitative spatial knowledge is pre-stored in long-term memory (LTM). Global path planner uses the spatial knowledge to plan a "fine-to-coarse" route. Then, the path information is temporarily stored in short-term memory (STM) and can be fetched by the goal generator to produce the next closest goal. The quantitative motion planning, occurred in working memory (WM). produces the motion behaviors to the closest goal. The robot current location information and contextual information in the movements is maintained by "I AM HERE" module. By re-planning the "fine-to-coarse" route during navigation, the robot always has a detailed closest goal to move till it reaches its destination. Finally, the proposed navigation system was demonstrated on the simulation platform and the pioneer3-DX robot.
|
PDF Full Text Request