Abstract
Localization is the problem of determining the position of a mobile robot from sensor data. Most existing localization approaches are passive, i.e., they do not exploit the opportunity to control the robot’s effectors during localization. This paper proposes an active localization approach. The approach provides rational criteria for (1) setting the robot’s motion direction (exploration), and (2) determining the pointing direction of the sensors so as to most efficiently localize the robot. Furthermore, it is able to deal with noisy sensors and approximative world models. The appropriateness of our approach is demonstrated empirically using a mobile robot in a structured office environment.
Introduction
Determining a robot's position from scratch without a known starting point—the "global localization" problem—is a fundamental challenge in robotics. In environments with high degrees of symmetry, such as long office corridors, traditional passive localization methods often fail. Passive approaches merely process incoming sensor data as the robot moves, often resulting in the robot wandering indefinitely without encountering unique features to fix its position. This research proposes that localization should be an active process where the robot intentionally chooses actions to "un-confuse" itself.
Active Localization Framework
The proposed framework is built upon Markov Localization, a probabilistic method that maintains a "belief" (a probability distribution) over all possible locations in a map. In symmetric environments, this belief often becomes multi-modal, meaning the robot thinks it could be in several identical-looking spots simultaneously.
To resolve this ambiguity, the system employs an Active Component guided by two primary strategies:
- Entropy Minimization: The robot calculates the "expected future entropy" for potential actions. Entropy serves as a mathematical measure of uncertainty; the robot selects actions predicted to result in the lowest future entropy (highest confidence).
- Active Navigation vs. Active Sensing:
- Active Navigation: The robot decides where to move by balancing the benefit of gaining certainty against the cost of travel using a value iteration cost function.
- Active Sensing: The robot determines where to look, firing specific sensors in a 360-degree array that are most likely to provide disambiguating information.
To maintain real-time performance, the authors utilize Position Probability Grids and approximate complex belief distributions using a small set of Gaussians, focusing computational resources on the most likely areas.
Experimental Results
The system was evaluated using RHINO, an RWI B21 mobile robot equipped with 24 sonar sensors, in a structured office environment at the University of Bonn.
Key findings include:
- Reliability: While passive methods (random wandering or wall-following) consistently failed in symmetric corridors, the active approach successfully localized the robot in all 20+ experimental trials.
- Efficiency: Active sensing resulted in a significantly faster drop in both entropy and actual estimation error compared to random sensing.
- Rational Behavior: The robot demonstrated the ability to seek out asymmetry. For example, when faced with two likely locations in a hallway, the robot would autonomously decide to enter a nearby room, knowing the room's unique layout would resolve the final ambiguity.
Conclusion
Active localization effectively transforms localization from a passive observation task into a goal-oriented control problem. By seeking out informative features rather than waiting to encounter them, robots can navigate complex, symmetric environments with much higher reliability. Future work will focus on optimizing the approach for massive environments and adapting the framework to handle dynamic obstacles like moving people.
Summary updated