🤖 AI Summary
This study addresses the challenge of unbounded spatial memory growth in visual robotic navigation within large-scale environments, which risks exhausting embedded platform resources. The authors conduct a systematic survey of 52 spatial memory representations from 1989 to 2025 and introduce, for the first time, a memory efficiency metric α—defined as the ratio of runtime memory to map storage—alongside a standardized evaluation protocol. Covering occupancy grids, neural implicit representations, 3D Gaussian Splatting (3DGS), and scene graphs, the analysis integrates GPU performance profiling and memory-completeness curves, revealing that neural methods exhibit α values spanning two orders of magnitude (2.3–215). This indicates that memory architecture, rather than representation paradigm, predominantly governs deployment feasibility. The work releases the first α benchmark dataset and provides α-aware budgeting algorithms with Pareto frontier analysis.
📝 Abstract
As vision-based robots navigate larger environments, their spatial memory grows without bound, eventually exhausting computational resources, particularly on embedded platforms (8-16GB shared memory, $<$30W) where adding hardware is not an option. This survey examines the spatial memory efficiency problem across 88 references spanning 52 systems (1989-2025), from occupancy grids to neural implicit representations. We introduce the $α= M_{\text{peak}} / M_{\text{map}}$, the ratio of peak runtime memory (the total RAM or GPU memory consumed during operation) to saved map size (the persistent checkpoint written to disk), exposing the gap between published map sizes and actual deployment cost. Independent profiling on an NVIDIA A100 GPU reveals that $α$ spans two orders of magnitude within neural methods alone, ranging from 2.3 (Point-SLAM) to 215 (NICE-SLAM, whose 47,MB map requires 10GB at runtime), showing that memory architecture, not paradigm label, determines deployment feasibility. We propose a standardized evaluation protocol comprising memory growth rate, query latency, memory-completeness curves, and throughput degradation, none of which current benchmarks capture. Through a Pareto frontier analysis with explicit benchmark separation, we show that no single paradigm dominates within its evaluation regime: 3DGS methods achieve the best absolute accuracy at 90-254,MB map size on Replica, while scene graphs provide semantic abstraction at predictable cost. We provide the first independently measured $α$ reference values and an $α$-aware budgeting algorithm enabling practitioners to assess deployment feasibility on target hardware prior to implementation.