Incremental Map-Making Based on Range Images

摘要

The problem for an operator to, without visual contact, estimate the posture of aremotely controlled vehicle has given birth to this work. The problem depends on the fact that current remote control systems only visualizes the vehicle's local environment of the vehicle, received from on-board TV cameras. The local environment is relevant for high precision operations, such as the manipulation of objects. For other operations, such as moving the vehicle or in navigational topics, the local environment is deficient. The weakness is compensated for by the introduction of a global environment, corresponding to a bird's-eye view containing the robot superimposed on the room in which it is working. Input to the global environment view consists of measurements by a laser scanner. As the scanner acts as a radar, some important objects may be hidden behind obstacles, giving a need for the merging of laser scans from different positions in the room. The problems thereby arising depend on the uncertainties in the laser data and in the vehicle position respectively. The former problem is solved by using a stochastic sensor model and representing sensor data in an Occupancy grid, while the latter is managed by using odometry along with a combination of Hough transforms, pattern matching, and Kalman filtering.