Vision-based urban navigation procedures for verbally instructed robots

Humans who explain a task to a robot, or to another human, use chunks of actions that are often complex procedures for robots. An instructible robot needs to be able to map such chunks to existing preprogrammed primitives. We investigate the nature of these chunks in an urban visual navigation context and describe the implementation of one of the primitives: "take the n^th turn right/left". This implementation requires the use of a "short-lived" internal map updated as the robot moves along. The recognition and localisation of intersections is done using task-guided template matching. This approach takes advantage of the content of human instructions to save computation time and improve robustness.