Decentralized Cooperative Localization for Multi-Robot Systems in Challenging Environments

Multi-robot systems provide more adaptability and robustness than conventional single-robot systems. This is because they have the ability to achieve partial completion of a task in the presence of failures due to the intrinsic inclusion of non-single-point failure. In multi-robot systems, cooperative localization refers to the use of relative measurements to improve the group’s overall localization performance. While centralized architectures in the system may result in unfeasible cases due to the high cost of computation and communication, decentralized methods distribute the computation among the robots in the group, enhancing the performance and making the system more efficient. Achieving a reliable localization performance is particularly challenging inGlobal Navigation Satellite System (GNSS) denied environments. Relying on GNSS only solutions for localization can suffer due to environmental restrictions since GNSS is often inaccessible or degraded for many locations, such as urban, indoor, subterranean, forested, and planetary environments. This thesis considers a decentralized cooperative localization algorithm that benefits from the use of different pseudo-measurement, such as Zero-Velocity Update (ZUPT) and Zero Angular Rate Update (ZARU), to improve the localization performance of the group. The Error-State Decentralized Extended Kalman Filter (ES-DEKF) utilizes proprioceptive (i.e., Inertial Measurement Unit (IMU), Wheel Encoders) and exteroceptive (i.e., Ultra-Wide Band (UWB), Camera) measurements to improve the localization performance of the system in a degraded GNSS environment. Multiple scenarios weretested in simulation and with real robotsto validatethe performance of the algorithm. Moreover, different heuristics of deciding when to utilize pseudo-measurements were compared to evaluate the benefit of these updates.