CS 326 A: Motion Planning Planning Exploration Strategies
Exploration Map Building A robot is introduced in a new environment Its task is to build a model (2D or 3D) of this environment It goes from place to place. At each place it acquires a local model that it merges with the current global model
Example
Example
Localization Issue Initial position of the robot reference coordinate frame Dead-reckoning (e.g., odometry) yields increasing errors Need to match each new local model with current global model (correspondence problem)
Correspondence Problem
Localization Issue Initial position of the robot reference coordinate frame Dead-reckoning (e.g., odometry) yields increasing errors Need to match each new local model with current global model Simultaneous Localization and Mapping (SLAM) Simultaneous Localization and Mapping (SLAM)
Loop Issue Courtesy R. Chatila Loops Complex optimization problem (incrementality vs. global optomization)
Not a New Issue!
Two Sub-problems in Exploration Where to go next? - Maximize amount of new information - Ensure minimal model overlap - Minimize travel length Next-best view How to best merge the two models? - Minimize expected errors: Kalman filtering, expectation maximization (most likely map)
Other Exploration Issues Map model: polygonal, occupancy grid, topological network, 3D surfaces, object map, etc… Moving objects “Meaning” of objects Multiple robots Unstructured environments
Other Exploration Issues Map model: polygonal, occupancy grid, topological network, 3D surfaces, object map, etc… Moving objects “Meaning” of objects Multiple robots Unstructured environments