Autonomous Navigation of a Known Map with a Segway Rmp

Yatima is a mobile robot on a Segway RMP base. This paper describes a method for autonomous navigation of a known map. The laboratory space Yatima is stored in is used as the navigation environment. A map is created, and the ROS navigation stack is used to implement autonomous navigation. Yatima can effectively navigate open areas, but struggles with obstacle avoidance. Simultaneous Localization and Mapping (SLAM) is a classic mobile robotics problem [1]. It focuses on the problem of mapping an unknown environment. The problem is typically set up by initializing a robot in an unknown location in an unknown environment. A solution to the SLAM problem will build a map of the environment as the robot moves as well as localize that robot within the map. This is typically a very difficult problem, as building an accurate map requires the precise location and orientation of the robot, and localizing the robot requires an accurate map. The key component of the SLAM problem is to estimate all previous poses of the robot and the map given the sensor history and odometry history. A robot " pose " consists of its location and orientation in the environment. The sensor history is a record of all sensor readings from the robot. The odometry reading is the record of all odometry information. The robot's odometry information provides data on how the robot is moving in the environment. It reports an estimated position and orientation for the robot based upon how the robot's drive base has been moving—for Yatima specifically, wheel rotation. More formally, the algorithm computes p(xt, m|z1:t, u1:t-1), where xt is the robot pose at time t, m is the map, z1:t is the observation history, and u1:t-1 is the odometry history. [1] Most SLAM algorithms use a particle filter to compute this probability. The specific implementation used by Yatima is GMapping, a SLAM algorithm using a Rao-Blackwellized filter developed by Giorgio Grisetti, Cyrill Stachniss, and Wolfram Burgard [1]. They describe the algorithm in detail in their paper, " Improved Techniques for Grid Mapping with Rao-Blackwellized Particle Filters. " The algorithm hinges on the factorization of the probability, p(xt, m|z1:t, u1:t-1) = p(m|x1:t, z1:t) * p(x1:t|z1:t, u1:t-1). They claim that the first probability of the factorization, p(m|x1:t, z1:t) can be computed analytically. The second probability is estimated using a particle filter. Each particle represents a singular robot pose at a given time. …