| The indoor mobile robots have broad application prospect, which can be used in service, entertainment, safe checking and so on. Indoor surroundings belong to complicated dynamic environment, which needs robot a higher surroundings sensing ability and responding ability, and the ability of tracking and avoiding moving object is most important. When running on dynamic environment such as home, it is inevitable for robot to be knocked by moving object or touched by a person, and the sensing and responding to bump is the basis for self-protection and being harmonious with environment. It is an important function for danger checking robot to detect the danger gas and fire indoor, which needs good performance of robot on danger environment sensing. The three environment sensing technologies are critical for indoor mobile robots when work on different condition.The relating aspects of the three environment sensing problem have been studied by researchers, but they are still not sufficient for specific application. The dissertation carries out theoretical and experimental research on the three problems for indoor mobile robot supported by National 863 Program.The HR-I indoor mobile robot system is designed for environment sensing studying, and the structure is optimized according to the demands of motion flexibility, support stability and structure simplification. The robot is designed with modular conception, on which many kinds of sensors are integrated to realize motion surroundings sensing, dangerous environment sensing and motion control. The control system is set up with hybrid architecture on the base of up and down computer control structure to improve the task managing ability and motion reacting ability.The motion environment sensing is mainly through Laser Range Finder, the sensing module is set up and the dynamic polar coordinate map model is presented referring to the grid map method based on the rolling window principle. The objects can be described with line and arc feature in the model. Line feature can be extracted with Recursive Line Fitting Method after the laser data is divided into segments to represent different objects. The obstacles are further described with three polar points, and the crowded obstacles will be merged into groups to reduce environment data.The tracking of moving obstacles is the basis of good obstacle avoiding performance. In order to realize obstacle tracking with moving platform, the position and posture estimating model of robot and its covariance model are set up based on odometer. Based on these models the moving obstacle tracking model is derived according to the Kalman Filter theory, which can estimate the current state and predict the future state of obstacle more accurately. To improve the obstacle avoiding flexibility the layered obstacle avoiding policy is adopted. For moving obstacle, an avoiding policy including waiting, bypassing and quickly going through is presented based on the prediction of collision time and position.The two-dimension accelerometer is adopted to perceive bump. The vibration signal caused by bump when robot is static is studied first to analyze the features of bump signal. According to the features of vibration signal resultant force and its direction, a rule based bump direction determing method is presented. Since the bumping direction is not very precise, a bump reacting policy according to different zone around robot is presented, which induce the robot make avoiding reaction or adjustment reaction. When robot is running, the velocity variation, the body vibration and road inclination can all induce acceleration signal, which will mix with bump signal. To recognize and extract the bump signal, a bump signal detection window is designed, which can work on time domain in real time. A dynamic bump responding approach is set up based on Motor Schema theory to consider obstacle and motion goal at the same time.In the process of danger environment detection, the gas source searching by robot is always a difficult problem. The responding and recovery delay of gas sensors are analyzed through experiment, and then the sensing model is also set up. For the environment that the danger sources are known, a Danger Site Directly Searching Policy is presented based on vision and gas sensors. For randomly distribute danger sources, the robot can judge the risk of suspected region through fuzzy logic by making use of the prior knowledge about gas distribution and source characters, and searching the region with higher risk grade step by step to find the source. Simulation shows the validity of the searching policy. Since gases can be released and the surrounding temperature will increase when there is a fire, the CO, CO2, CH4 and temperature sensors are used to detect fire, and a fire intensity estimating model is set up based on Support Vector Machine theory to judge the fire level. The contrast to neural network shows better classification and generalization ability.The simulation and experiment show the validity of theory approach, the robot can make good sensing to its environment and respond to it with better performance.
|
PDF Full Text Request