Distributed object based SLAM

Abstract

The use of multiple cooperative robots or mobile devices has the potential to enable fast information gathering, and more efficient coverage and monitoring of large areas. In particular, distributed SLAM, i.e., the cooperative construction of a model of the environment explored by the robots or mobile devices, is fundamental to geotag sensor data (e.g., for pollution monitoring, surveillance and search and rescue), and to gather situational awareness. For military applications, multi-robot systems promise more efficient operation and improved robustness to adversarial attacks. In civil applications, the use of several inexpensive, heterogeneous, agile platforms is an appealing alternative to monolithic single robot systems.

In this thesis, we aim at designing a technique that allows each robot or mobile device to build its own object level map while asking for minimal knowledge of the map of the teammates. In particular, we make the following three major contributions:

1. We present a distributed algorithm based on Distributed Gauss-Seidel to estimate the 3D trajectories of multiple cooperative robots from relative pose measurements. This approach has several advantages. It requires minimal information exchange, which is beneficial in presence of communication and privacy constraints. It has an anytime flavor: after few iterations, the trajectory estimates are already accurate, and they asymptotically convergence to the centralized estimate. The DGS approach scales well to large teams, is resistant to noise and it has a straightforward implementation. We test the approach in simulations and field tests, demonstrating its advantages over related techniques.

2. We present an approach for distributed SLAM which uses object landmarks in a distributed mapping framework. We show that this approach further reduces the information exchange among robots (as compared to feature based DGS), results in a compact, human understandable map, and has lower computational complexity as compared to low-level feature based distributed mapping.

3. Finally, we extend the previous work to the case where object models are previously unknown and are modeled jointly with Distributed Object-based SLAM. We show that this approach further reduces the memory required to store the object models while maintaining the accuracy at the same level as the state of art RGB-D mapping approaches.