NAVIGATION AND MAPPING WITHIN THE CONSTRAINTS OF A MARS MICRO-ROVER

摘要

Planetary rover missions have thus far performedin relatively benign environments to prevent loss of expensivehardware in risky, but potentially more scientifically valuableenvironments. Mission risk can be reduced by using micro-roverscouts that map the terrain prior to the main mission. However,micro-rovers will have more limited power and computationability than their larger counterparts. Kapvik, a Canadian microrover,shares these power and computation constraints. Thispaper explores methods of implementing simultaneous localizationand mapping (SLAM) using a laser range finder andstereo cameras on Kapvik along with inertial sensors includinggyroscopes, accelerometers, a sun sensor and wheel encoders.To perform SLAM in outdoor and unstructured environments 6degree-of-freedom SLAM is usually required, where the completepose of the rover is estimated along with the 3D position of itsenvironment. This is a computationally expensive problem torun in an online fashion. The FastSLAM algorithm has beenselected for Kapvik's navigation due to its effciency, scalability,and robustness. The performance of a novel modification ofthe FastSLAM algorithm using the nonlinear Cubature KalmanFilter is compared with other implementations of the FastSLAMalgorithm. The modification also makes use of the inertial sensorsaboard Kapvik to improve its localization between scans of itsenvironment with either the laser or camera. By increasing thepose estimation accuracy between scans the number of times laseror camera data must be processed is reduced. For initial testingof the SLAM algorithms a simulation environment is developed.This includes a 3D motion model for the Kapvik micro-roverwhich incorporates an articulated rocker-bogie chassis drivingover uneven terrain including wheel slip as an external input.Mapping sensors such as a laser range finder and the inertialsensors that sense the rover's motion are also simulated inaddition to terrain being randomly generated that representsa planetary surface for the rover to traverse. Finally, theimplementation of the algorithm on actual Kapvik hardware isoutlined and the process to validate the results is discussed. Thestandard operation of the rover's navigation system including theautonomous decision process to take a scan as it drives along aspecifed path is also presented. Testing using the Kapvik microroverdemonstrates the importance of Earth analogue missionsas precursors to missions to Mars.