Navigation By Path Remembering

摘要

We present a scheme for navigation that does not rely on a detailed world map or a precise inertial guidance system. Our approach is based on a multiple-agent control system in which there are a variety of local navigation behaviors that run in parallel. These behaviors arbitrate among themselves to select the most applicable action in any given situation. Meanwhile, there is another process which watches for "choice points" -- places where it is possible to travel in one of several distinct directions -- and simply records the qualitative direction actually travelled at each point. Our representation of the robot's path is then a sequence of these choices which can be replayed in reverse to get the robot back to its starting location. An important point is that instead of attempting to build a complete internal map, we use the world as its own representation. Another feature is that the robot does not even care which algorithm(s) are employed to get it from one point to the next. The individual agents communicate through the world to the path rememberer by turning the robot.