Reliable navigation using landmarks

Building a truly reliable mobile robot system-one that can navigate without failures for long periods of time (weeks or months)-requires making clear assumptions bounding uncertainty and enforcing those assumptions by appropriately engineering the robot and its workspace. Weak assumptions may result in low-cost engineering but make the navigation problem intractable. On the other hand, strict assumptions may simplify navigation but reduce the flexibility of the resulting system. The work presented in this paper investigates the tradeoff between "computational complexity" and "physical complexity" and advocates landmarks as a way of managing this tradeoff. We first define a formal navigation problem which incorporates enough assumptions to make it computationally tractable. We then use landmarks to enforce those assumptions. By implementing this system on our mobile robot we show that the assumptions are enforceable, that the engineering costs of using landmarks are acceptable, and that the resulting navigation system is both efficient and robust.