Multi Robot Object-based SLAM Siddharth Choudhary 1 , Luca Carlone 2 , Carlos Nieto 1 , John Rogers 3 , Zhen Liu 1 , Henrik I. Christensen 1 , and Frank Dellaert 1 1 Institute for Robotics and Intelligent Machines, Georgia Institute of Technology 2 Laboratory for Information and Decision Systems, Massachusetts Institute of Technology 3 Army Research Lab Abstract. We propose a multi robot SLAM approach that uses 3D ob- jects as landmarks for localization and mapping. The approach is fully distributed in that the robots only communicate during rendezvous and there is no centralized server gathering the data. Moreover, it leverages local computation at each robot (e.g., object detection and object pose estimation) to reduce the communication burden. We show that object- based representations reduce the memory requirements and information exchange among robots, compared to point-cloud-based representations; this enables operation in severely bandwidth-constrained scenarios. We test the approach in simulations and field tests, demonstrating its advan- tages over related techniques: our approach is as accurate as a centralized method, scales well to large teams, and is resistant to noise. 1 Introduction The deployment of multiple cooperative robots is an important asset for fast information gathering over large areas. In particular, multi robot SLAM, i.e., the cooperative construction of a model of the environment explored by the robots, is fundamental to geo-tag sensor data (e.g., for pollution monitoring, surveillance and search and rescue), and to gather situational awareness. In this paper we are interested in the case in which the robots operate under severe bandwidth constraints. In this context, the robots have to reconstruct a globally-consistent map by communicating a small amount of information among the teammates. Dealing with bandwidth constraints is challenging for two reasons. First, most approaches for multi robot SLAM imply a communication burden that grows quadratically in the number of locations co-observed by the robots [1]; these approaches are doomed to quickly hit the bandwidth constraints. In our previous works [2, 3] we alleviated this issue by proposing an approach, based on the distributed Gauss-Seidel method, which requires linear communication. The second issue regards the communication cost of establishing loop closures among robots. When the robots are not able to directly detect each other, loop closures have to be found by comparing raw sensor data; in our setup the robots are equipped with an RGBD camera and exchanging multiple 3D point clouds