Research Of Autonomous Navigation For The Mobile Agent In Unknown Environment

Research of the autonomous mobile agent is an important field for the research and engineering application of the agent. For a mobile agent, its ability of autonomous navigation is an important part of its various capacities. Autonomous navigation for the mobile agent includes online path planning, obstacle avoidance, localization of the mobile agent, and environment and obstacle perception, etc. It requires the mobile agent with no prior information of environment or obstacles, only perceiving surrounding environment information according to its own perception equipment, and making the mobile agent autonomously plan its movement and moving trajectory based on its sensing information. Eventually, the mobile agent can autonomously move from the starting position (or initial configuration) to the specified goal position (or final configuration) as efficient and reliable as possible. Autonomous navigation for the mobile agent requires the agent meeting certain constraints, such as collision-free with the obstacle, the shortest path length, the lowest energy consumption, the shortest time cost, and kinematic constraints, etc.This paper studies the autonomous navigation problem for the mobile agent in unknown environment within the scope of large-scale and small-scale. Major researches and innovations are summarized as follows: 1. The concept of vectorization path is proposed in this research, and the planned path is described in the form of vector set. All the planned paths are described and saved in a set of end-to-end vectors. Different from other path representation, when the mobile agent moves along the vectorization path, it can directly execute the moving and steering instructions according to the planning path. And the proposed method can effectively avoid the dependence on map construction in the planning process. There is no huge data requirement for map storage and maintenance, so the space requirement for path saving can be reduced significantly. And the proposed path regeneration method can facilitate the maintenance and updating of the path, and greatly reduce the amount of computation on path updating.2. This research combines the vectorization path representation and the thinking of Bug algorithm in order to solve the autonomous navigation problem for the mobile agent in static unknown environment within the scope of large-scale. The study draws on the move-to-goal and boundary-following thinking in Bug family, and combines them firmly with the vectorization path thinking. Firstly, construct the initial path in accordance with the starting position and the goal position. And then make the mobile agent moving along the initial path. In the process of moving, the mobile agent perceives the obstacles in the completely unknown environment depending on the different conditions. Once there is obstacle constructing the current path, an intermediate point will be generated and inserted into the current path to get a regenerated collision-free path. And then make the mobile agent move along the regenerated path, so as to arrive at the goal position with no collision. In this research, two intermediate point generation methods are proposed, which are scanned generating method and randomly generating method. Compare to random generating method, scanned generating method can get a shorter path length, but random generating method has better global search ability. The shortcoming of scanned generating method is that the mobile agent is easily getting the local optimal solution. The research proposed improving strategy for this shortcoming in order to make the agent quickly out of the local optimum after it falling into the local optimum. For the random generating method, the rapid convergence strategy is proposed to get shorter path trajectory and executing time.3. The vectorization path representation is applied to the research of autonomous navigation for mobile agent in dynamic unknown environment within the scope of large-scale, in order to solve the path planning problem in the environment with moving obstacles. According to the vectorization path representation thinking, establish the initial path firstly, and make the mobile agent moving along the path with perceiving the environment at certain time interval. Once the obstacle is found, the mobile agent will determine the moving trend of the obstacle, and make a judgment of whether the obstacle will construct the current path or not. If there is or will be an obstacle constructing the current path, an intermediate point, in the opposite direction to the moving direction of obstacle, will be generated in accordance with the proposed strategy. And then the collision-free path will be regenerated by inserting the intermediate point. The mobile agent will move along the regenerated path to arrive at the goal position collision-free.4. Taking full consideration of the kinematic models and constraints of the mobile agent, a novel autonomous navigation method for mobile agent in the range of small-scale based on Bug thinking is proposed. Combining the Dubins path with the Bug family algorithm, when the mobile agent moving from one configuration to another, it always moving along the Dubins path which is made up in accordance with the agent’s minimum turning radius, in order to meet the kinematic constraints of the mobile agent. First of all, make the mobile agent moving from the initial configuration to the final configuration along the Dubins path. Perceive the obstacles in environment under certain conditions while moving. If there is obstacle constructed the current path, generate the intermediate configuration in accordance with the proposed strategy, and make the mobile agent move along the new Dubins path from current configuration to intermediate configuration. The Dubins path from intermediate configuration to the final configuration is also generated. Repeat the series of process, such as moving, perception and intermediate configuration generation, until the mobile agent arrives at the final configuration with no collision.