Decentralized Collaborative Navigation with Enhanced Positioning Update for Mobile Robots

Collaborative navigation or shared localization techniques are employed to enhance self-localization accuracy through information exchange among multiple agents operating together. This improvement is achieved by fusing sensor data from various sources, which can originate from multiple agents. Most existing approaches share information about the environment itself such as 3D modeling or SLAM and are considered to be computationally expensive, since they require high bandwidth communication system to enable such information sharing about the environment. Such high bandwidth communication system are not always available, making such information sharing not feasible. Alternative approaches use both range and angular data from high-end radar sensors. However, to reduce size, power consumption, and weight, the use of high-end radars should be minimized in favor of using only range measurement sensors. In this work, we address the self-localization problem for an agent operating independently within an environment containing additional agents. The agent is equipped with commonly used sensors, including a GNSS receiver, INS, a low-bandwidth communication system, and a low-end range sensor that provides only range information. Rather than relying solely on GNSS position updates, we propose constructing two enhanced position update methods for the agent's navigation filter, integrating range measurements and communication with other agents. Unlike existing approaches, we utilize range measurements exclusively, without any angular information. Our problem formulation adopts a decentralized approach, making it well-suited for automotive applications where the environment is highly dynamic and scalability with the number of nearby agents is crucial. We compare our proposed approach using several fusion methods in simulation and prove our results in a real lab experiment including multi-agents. Our results demonstrate that our rangeonly measurement approach achieves comparable performance to methods that utilize both range and angular information. An example video from one of our experiments can be found here.