D-Lite: Navigation-Oriented Compression of 3D Scene Graphs for Multi-Robot Collaboration

TL;DR

This work tackles efficient map sharing for multi-robot exploration by compressing 3D Scene Graphs under strict communication budgets while preserving navigation-relevant information. It introduces D-Lite, a graph-spanner–driven framework with two greedy algorithms, BUD-Lite (bottom-up) and TOD-Lite (top-down), that exploit DSG hierarchy to balance data size and path distortion. The approach is backed by both a spanner-based problem formulation and real-time algorithms, with simulations showing that navigation performance degrades minimally (e.g., up to 8% extra time) when transmitting as little as 1.6% of the original DSG. The methods offer practical, scalable, and task-driven compression for robust multi-robot collaboration in unknown environments, and point to extensions for dynamic settings and real-world robotic validation.

Abstract

For a multi-robot team that collaboratively explores an unknown environment, it is of vital importance that collected information is efficiently shared among robots in order to support exploration and navigation tasks. Practical constraints of wireless channels, such as limited bandwidth, urge robots to carefully select information to be transmitted. In this paper, we consider the case where environmental information is modeled using a 3D Scene Graph, a hierarchical map representation that describes both geometric and semantic aspects of the environment. Then, we leverage graph-theoretic tools, namely graph spanners, to design greedy algorithms that efficiently compress 3D Scene Graphs with the aim of enabling communication between robots under bandwidth constraints. Our compression algorithms are navigation-oriented in that they are designed to approximately preserve shortest paths between locations of interest, while meeting a user-specified communication budget constraint. The effectiveness of the proposed algorithms is demonstrated in synthetic robot navigation experiments in a realistic simulator. A video abstract is available at https://youtu.be/nKYXU5VC6A8.

Figures (17)

• Figure 1: 3D Scene Graph of an environment (left) and compressed version produced by D-Lite (right). The purple circles mark the terminal nodes: D-Lite approximately preserves shortest-path distances between those locations of interest.
• Figure 2: Distortion of shortest path from $s$ to $t$ (thick red).
• Figure 3: Illustration of the BUD-Lite procedure with source $s$ and targets $t_1, t_2$. At each iteration, place nodes in a shortest path between terminals are replaced by a room node. Nodes are removed when none of the terminal pairs $(s,t_1)$ and $(s,t_2)$ connects through them. Note that the final graph cannot be further pruned without disconnecting terminals.
• Figure 4: Initial (left) and final DSG (right). Terminal nodes (A, B, C, and D) are in blue, place nodes in red, and the room node in green.
• Figure 5: Illustration of the TOD-Lite expansion procedure with one source $s$ and two targets $t_1$ and $t_2$. At each iteration, a room node is expanded and replaced with its children place nodes. Adjacent nodes may be added to ensure connectivity (e.g.,$P_3$ at first iteration).

Theorems & Definitions (5)

• Definition 1: Ancestor
• Definition 2: Diameter
• Proposition 1: Worst-case BUD-Lite stretch
• proof
• proof