Edge-Assisted Multi-Robot Visual-Inertial SLAM with Efficient Communication

Xin Liu; Shuhuan Wen; Jing Zhao; Tony Z. Qiu; Hong Zhang

cs.ROMar 11, 2026

Edge-Assisted Multi-Robot Visual-Inertial SLAM with Efficient Communication

Authors: Xin Liu, Shuhuan Wen, Jing Zhao, Tony Z. Qiu, Hong Zhang

Organizations: the Department of Key Lab of Industrial Computer Control Engineering of Hebei Province, Engineering Research Center of the Ministry of Education for Intelligent Control System and Intelligent Equipment, Yanshan University, Qinhuangdao, 066004, China. · Intelligent Transport System Research Center, Wuhan University of Technology, Wuhan, China · the Department of Civil and Environmental Engineering at the University of Alberta, Canada. · the Department of Electronic and Electrical Engineering, Southern University of Science and Technology, Shenzhen, China.

Abstract

The integration of cloud computing and edge computing is an effective way to achieve global consistent and real-time multi-robot Simultaneous Localization and Mapping (SLAM). Cloud computing effectively solves the problem of limited computing, communication and storage capacity of terminal equipment. However, limited bandwidth and extremely long communication links between terminal devices and the cloud result in serious performance degradation of multi-robot SLAM systems. To reduce the computational cost of feature tracking and improve the real-time performance of the robot, a lightweight SLAM method of optical flow tracking based on pyramid IMU prediction is proposed. On this basis, a centralized multi-robot SLAM system based on a robot-edge-cloud layered architecture is proposed to realize real-time collaborative SLAM. It avoids the problems of limited on-board computing resources and low execution efficiency of single robot. In this framework, only the feature points and keyframe descriptors are transmitted and lossless encoding and compression are carried out to realize real-time remote information transmission with limited bandwidth resources. This design reduces the actual bandwidth occupied in the process of data transmission, and does not cause the loss of SLAM accuracy caused by data compression. Through experimental verification on the EuRoC dataset, compared with the current most advanced local feature compression method, our method can achieve lower data volume feature transmission, and compared with the current advanced centralized multi-robot SLAM scheme, it can achieve the same or better positioning accuracy under low computational load.

Figures & tables

Panel of Figure 1 — Fig. 1: Multi-robot Collaborative SLAM communication architecture.

Fig. 2: System overview. Each robot is equipped with a Visual-Inertial sensor suite. After analyzing the sensor information, the compressed data is transmitted to a fixed edge server. The edge server can conduct location and map construction in local scenes by running the VIO in real-time and independently. The communication interface is used to exchange data (keyframe local binary features and mappoints) between the edge server and the cloud. The cloud can run computationally expensive, global, and non-real-time tasks: redundant keyframe culling, loop closure, and global map fusion.

Fig. 3: IMU pre-integration diagram. Integrating all IMU information between the $k\rm{th}$ frame and the $(k+1)\rm{th}$ frame, the position, velocity and rotation (PVQ) of the $(k+1)\rm{th}$ frame can be obtained as the initial value of the visual estimation.

Fig. 4: The choice of encoding mode. The decision of keyframes depends on the number of points tracked by the IMU-assisted LK optical flow and the time interval from the previous keyframe.

Panel of Figure 5 — Fig. 5: Visual-Inertial joint optimization factor graph.

Fig. 6: Visual-Inertial joint optimization factor graph in the cloud.

Panel of Figure 7 — Fig. 7: The tracking results of traditional LK optical flow tracking and IMU-assisted optical flow tracking on EuRoC. Raw image data (a), traditional LK optical flow tracking results without IMU participation (b), IMU-assisted LK optical flow tracking results (c).

Fig. 8: The number of points tracked and the track time spent. The blue bar represents the number of points tracked by LK optical flow method, and the orange bar represents the number tracked by the proposed. The purple broken line represents the time spent in LK tracking and the green broken line represents the time spent in the proposed tracking.

Methods	Intra-Coding in [ 58 ]	The Proposed
		KF	Non-KF
Memory	217 bits	217 bits	27 bits

TABLE I: Comparison of the memory required for a feature point

Fig. 9: The proportion of each part of each feature in different methods under lossless compression.

		Sequences
		MH_01-MH_03	MH_01-MH_05	V1_01-V1_03	V2_01-V2_03
Method	CCM-SLAM [ 49 ]	0.077	-	-	-
	VINS-mono [ 33 ]	-	0.210	-	-
	ORB-SLAM3 [ 15 ]	0.037	0.065	0.040	0.048
	CVIDS [ 51 ]	-	0.074	-	-
	COVINS AWS [ 11 ]	0.025	0.039	0.049	-

TABLE II: RMSE of ATE in units of on the EuRoC dataset. Our method is to calculate the average after five runs. Note that the bold means the best, and ”-” means that the result of the corresponding method cannot be found in the literature.

Methods	ATE/m	Number of Keyframes	Number of Mappoints
The Proposed (w/o MBP)	0.051	1550	57169
The Proposed (w/ MBP)	0.046	925	41923

TABLE III: Comparison of results using different redundant keyframe removal methods.

Panel of Figure 13 — Fig. 10: The results of collaborative localization and mapping in the V2 sequences of the proposed method.

Panel of Figure 14 — Fig. 11: The trajectories of collaborative localization and mapping in the V2 sequences of the proposed method.

Fig. 12: The experimental scene and the equipments.

Panel of Figure 16 — Fig. 13: The results of collaborative localization and mapping in the real scene of the proposed method.

Fig. 14: Comparison of measured trajectory and groundtruth in real scene.

Fig. 15: Trajectory overlaid with 2D grid map of the real scene for visual comparison.

Shuhuan Wen was received the PhD degree in control theory and control engineering from the Yanshan University, Qinhuangdao, China in 2005. She is currently a Professor of automatic control in the Department of Electric Engineering, Yanshan University. She has coauthored one book, about 40 papers. Her research interests include SLAM, computer vision, robotics, 3D object recognization and reconstruction. Dr. Wen was a Visiting Scholar of the Ottawa University, Carleton University and Simon Fraser University in Canada in 2011-2013. She is also a visiting professor at University of Alberta in Canada in 2021-2022. She is an AE of IROS in 2021 to 2022.

Table 20

Hong Zhang was received the B.Sc degree from Northeastern University, Boston, USA, in 1982, and the PH.D.degree from Purdue University. West Lafayette, IN, USA, in 1986, both in electrical engineering. He conducted post-doctoral research at the University of Pennsylvania from 1986 to 1987 before joining the Department of Computing Science, University of Alberta, Canada. He is an NSERC Industrial Research Chair. His current research interests include robotics computer vision, and image processing.

Table 23

Simultaneous Localization and Mapping

Simultaneous localization and mapping estimates a robot’s motion while building a map of an unknown environment.

See all 197 papers

riMESA: Consensus ADMM for Real-World Collaborative SLAM
Mar 1, 2026Daniel McGann, Michael KaessSimultaneous Localization and Mapping Pose Graph Optimization
The authors are with the Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA.
SGM-SLAM: Scene Graph Matching for Data-Efficient Distributed SLAM
Jun 15, 2026Yewei Huang, Tixiao Shan, Abhinav Rajvanshi +4Simultaneous Localization and Mapping Multi-Robot Systems
Dartmouth College, 100 N Main St, Hanover, NH 03755, USA · SRI International, 201 Washington Rd, Princeton, NJ 08540, USA · Stevens Institute of Technology, 1 Castle Point Terrace, Hoboken, NJ 07030, USA
Above and Below: Heterogeneous Multi-robot SLAM Across Surface and Underwater Domains
May 10, 2026John McConnell, Armon Shariati, Paul Szenher +1Simultaneous Localization and Mapping Multi-Robot Systems
United States Naval Academy, Annapolis, MD, USA · Shield-AI, San Diego CA, USA · Stevens Institute of Technology, Hoboken, NJ, USA

Mar 1, 2026J/K move · Enter open · S save

riMESA: Consensus ADMM for Real-World Collaborative SLAM

Daniel McGann, Michael Kaess

The authors are with the Robotics Institute, Carnegie Mellon University, Pittsburgh, PA, USA.

Simultaneous Localization and Mapping Pose Graph Optimization Alternating Direction Method of Multipliers Distributed Optimization

Open paper AskPDF arXiv

Collaborative Simultaneous Localization and Mapping (C-SLAM) is a fundamental capability for multi-robot teams as it enables downstream tasks like planning and navigation. However, existing C-SLAM back-end algorithms that are required to solve this problem struggle to address the practical realities of real-world deployments (i.e. communication limitations, outlier measurements, and online operation). In this paper we propose Robust Incremental Manifold Edge-based Separable ADMM (riMESA) -- a robust, incremental, and distributed C-SLAM back-end that is resilient to outliers, reliable in the face of limited communication, and can compute accurate state estimates for a multi-robot team in real-time. Through the development of riMESA, we, more broadly, make an argument for the use of Consensus Alternating Direction Method of Multipliers as a theoretical foundation for distributed optimization tasks in robotics like C-SLAM due to its flexibility, accuracy, and fast convergence. We conclude this work with an in-depth evaluation of riMESA on a variety of C-SLAM problem scenarios and communication network conditions using both synthetic and real-world C-SLAM data. These experiments demonstrate that riMESA is able to generalize across conditions, produce accurate state estimates, operate in real-time, and outperform the accuracy of prior works by a factor >7x on real-world datasets.

Abstract

Simultaneous Localization and Mapping

Explore similar work

riMESA: Consensus ADMM for Real-World Collaborative SLAM

SGM-SLAM: Scene Graph Matching for Data-Efficient Distributed SLAM

Above and Below: Heterogeneous Multi-robot SLAM Across Surface and Underwater Domains