Grid Cell-Inspired Fragmentation and Recall for Efficient Map Building
Jaedong Hwang, Zhang-Wei Hong, Eric Chen, Akhilan Boopathy, Pulkit, Agrawal, Ila Fiete

TL;DR
This paper introduces FARMap, a novel map-building approach inspired by neuroscience, which uses fragmentation and recall mechanisms to efficiently explore and map large environments by building local maps and recalling them based on surprisal-driven events.
Contribution
The paper proposes a new fragmentation-and-recall mapping method inspired by grid cell behavior, improving efficiency in large environment mapping.
Findings
FARMap replicates animal-like fragmentation points.
It covers environments faster than existing methods.
It uses less active memory without performance loss.
Abstract
Animals and robots navigate through environments by building and refining maps of space. These maps enable functions including navigation back to home, planning, search and foraging. Here, we use observations from neuroscience, specifically the observed fragmentation of grid cell map in compartmentalized spaces, to propose and apply the concept of Fragmentation-and-Recall (FARMap) in the mapping of large spaces. Agents solve the mapping problem by building local maps via a surprisal-based clustering of space, which they use to set subgoals for spatial exploration. Agents build and use a local map to predict their observations; high surprisal leads to a "fragmentation event" that truncates the local map. At these events, the recent local map is placed into long-term memory (LTM) and a different local map is initialized. If observations at a fracture point match observations in one of the…
Peer Reviews
Decision·Submitted to ICLR 2024
The use of a non-uniform submap generation logic is interesting.
While the idea of surprise-based submap creation is interesting, many aspects of the overall method are unclear. What does the map representation look like? The information provided appears to be contradicting itself. The C-th channel is said to contain confidence information, but over what? Additionally, the C-th channel in the observation contains visibility information. However, later on, there is talk of occupancy and colors. The actual representation used by the maps is never explained con
- the paper overall presents a technically sound method that is able to achieve exploration of unknown environments. - the paper provides an interesting grounding of the proposed method with neuroscience, in proposing fragmentation and recall. - the paper is overall clear, with a logical structure in presenting the different components of the proposed method.
- while it is interesting to see the grounding of the proposed method in neuroscience, some of the general ideas are already present in other methods for exploration, in particular, reasoning topologically is captured by methods that use the generalized Voronoi graph or semantic maps to guide the exploration, and the long-term storage through pose graphs in SLAM, where loop closure is applied (discussed in graph-based slam appendix section), or curiosity-driven exploration. The paper should disc
— the paper is well written. — using neurophysiological knowledge of the rodent hippocampus to inform the design of spatial navigation system is interesting — the results seem to be promising
— the improvement over the alternative methods seems to shrink in the more real-world-like applications. Can the authors comment on or provide an interpretation of this? — Can the authors justify why the Frontier method by Yamauchi (1997) would be the most appropriate benchmark to have? It would be nice if it’s possible to include some other more recent methods.
Code & Models
Videos
Taxonomy
TopicsRobotics and Sensor-Based Localization · Robotic Path Planning Algorithms · Advanced Image and Video Retrieval Techniques
MethodsFragmentation
