Trail-Map: A scalable landmark data structure for biologically inspired range-free navigation

Small mobile robots often have very limited computational resources, but should still be able to navigate robustly in spacious unknown environments. While local navigation already requires high-resolution maps for obstacle avoidance and path planning, the global navigation task should aim to consume as little resources as possible but still enable the robot to robustly find the way to important places. Inspired by models of insect navigation, Augustine et al. [1] recently introduced the LT-Map, a scalable data structure for homing based on bearing-only landmark measurements. The robot memorizes landmark configurations during its first traversal of a path and uses them to navigate the same route again. The LT-Map uses a tree structure to store the landmark views in the order of their translation invariance. This paper introduces an improvement of the LT-Map, the Translation Invariance Level Map (Trail-Map). This novel data structure also stores the landmark views in a hierarchical order of translation invariance, but is based on lists of landmark views. Thus, it avoids redundancies that could arise in the LT-Map and leads to a more consistent hierarchy. The Trail-Map achieves significant memory savings and can be created and pruned very efficiently what makes it attractive to mobile robots with limited computational power. Simulation results show that the Trail-Map data structure can save more than 80% of memory compared to the LT-Map while achieving the same path accuracy.