Abstract
Traditional robotic navigation often prioritizes high-precision metric mapping, a stark contrast to the fragmentary, topological "cognitive maps" used by biological organisms. This paper presents a shift toward biologically-inspired navigation using the Absolute Space Representation (ASR) model. By fusing imprecise sonar data and odometry through a dynamic weighting algorithm known as Democratic Integration, a mobile robot successfully navigates and localizes within an office environment without the need for a pixel-perfect global map.
Introduction
In the field of robotics, Simultaneous Localization and Mapping (SLAM) typically aims for mathematical perfection, attempting to represent every surface with millimeter precision. However, humans and animals navigate effectively using a sequence of "places"—rooms, corridors, and junctions—rather than absolute coordinates. This research, presented at IJCAI-07, explores whether a robot can find its way home using an intentionally imprecise, fragmentary network of local spaces, mimicking early-stage biological spatial learning.
Methodology: The ASR Model
The researchers utilize the Absolute Space Representation (ASR) model, which treats the environment as a network of discrete local spaces.
Mapping and Segmentation
A Pioneer 2 robot equipped with sonar sensors explores the environment. Instead of maintaining a global grid, it employs a "split and merge" algorithm to segment its path into distinct ASRs, such as a specific stretch of a corridor or a room.
Localization Heuristics
To navigate back to its starting point, the robot relies on two "weak" heuristics:
- ASR-Distance: The robot measures the distance between the entrance and exit of each local space. This abstracts away minor obstacle-avoidance maneuvers, providing a stable metric for the "length" of a space.
- Relative Orientation: The robot records the sequence of turns (angles) between adjacent ASRs, creating a unique "fingerprint" for its location.
Democratic Integration
The core of the sensor fusion is "Democratic Integration." Rather than hard-coding trust levels, the robot uses a recursive weighting system. If a heuristic (e.g., distance) provides high-confidence results consistent with the global estimate, its weight increases. If it becomes unreliable due to factors like wheel slippage, its influence is automatically reduced.
Experimental Results
The approach was evaluated through real-world experiments in an office environment featuring corridors, doors, and dynamic obstacles.
Successful Localization
The robot consistently identified its current ASR during return journeys. The system demonstrated robustness even when the environment changed slightly (e.g., moved rubbish bins) between the outbound and inbound trips.
Dynamic Weight Adaptation
Results confirmed that the robot dynamically adjusts its reliance on different cues. While distance metrics were often favored initially, the system shifted weight toward orientation data as cumulative odometry errors increased or when moving through complex transition regions.
Discussion and Conclusion
This research demonstrates that high-level navigation does not strictly require high-precision metric maps. By focusing on "local space indices" rather than exact (x, y) coordinates, robots can achieve reliable navigation using low-fidelity sensors.
Future work may address the "Turn Around" problem—how a robot re-orients itself at the start of a return journey—and the integration of visual landmarks to further enhance the robustness of these cognitive maps in more complex, open environments.
Summary updated