Gas source localization (GSL) by an olfactory robot is a research field with a great potential for applications but also with numerous unsolved challenges, particularly when the search must take place in realistic, indoor environments that feature obstacles and turbulent airflows. In this work, we present a new probabilistic GSL method for a terrestrial mobile robot that revolves around the propagation of local estimations throughout the environment. By exploiting the geometry of the environment as the basis for this propagation, we avoid relying on analytical dispersion models, eliminating the need to assume controlled environmental conditions. Simulated and real experiments are presented in different indoor environments featuring multiple rooms and turbulent flows, demonstrating the suitability of our approach for locating the emitting gas source.

Gas source localization ( GSL ) with an autonomous robot is a problem with many prospective applications, from finding pipe leaks to emergency-response scenarios. In this work, we present a new method to perform GSL in realistic indoor environments, featuring obstacles and turbulent flow. Given the highly complex relationship between the source position and the measurements available to the robot (the single-point gas concentration, and the wind vector) we propose an observation model that derives from contrasting the online, real-time simulation of the gas dispersion from any candidate source localization against a gas concentration map built from sensor readings. To account for a convenient and grounded integration of both into a probabilistic estimation framework, we introduce the concept of probabilistic gas-hit maps, which provide a higher level of abstraction to model the time-dependent nature of gas dispersion. Results from both simulated and real experiments show the capabilities of our current proposal to deal with source localization in complex indoor environments.

The simulation of how a gas disperses in a environment is a necessary asset for the development of olfaction-based autonomous agents. A variety of simulators already exist for this purpose, but none of them allows for a sufficiently convenient integration with other types of sensing (such as vision), which hinders the development of advanced, multi-sensor olfactory robotics applications. In this work, we present a framework for the simulation of gas dispersal and sensing alongside vision by integrating GADEN, a state-of-the-art Gas Dispersion Simulator, with the Unity 3D, a video game development engine that is used in many different areas of research and helps with the creation of visually realistic, complex environments. We discuss the motivation for the development of this tool, describe its characteristics, and present some potential use cases that are based on cutting-edge research in the field of olfactory robotics.

There are many potential applications for an autonomous robotic agent capable of sensing gases in the environment, from locating leaks in pipes to monitoring air quality. However, the current state of the art in the field of robotic olfaction is not mature enough for most real-world applications. Due to the complexity of gas dispersion phenomena and the limitations of sensors, a great deal of research into the development of techniques and algorithms remains necessary. A very important part of this research is thorough experimentation, but carrying out robotic olfaction experiments is far from trivial. Real world experiments are usually limited to very simplified, wind-tunnel-like environments, as it is impossible to closely monitor or control the airflow in more complex scenarios. For this reason, simulation with CFD offers the most plausible alternative, allowing researchers to study the behavior of their algorithms in more challenging and complex situations. This work presents a CFD-based gas dispersion dataset composed of 120 cases generated under variable environmental conditions, taking place in 30 realistic and detailed models of real houses. All the data is made available in multiple formats, and is directly accessible through ROS, to permit easy integration with other robotic tools.

Robots in human-centered environments require accurate scene understanding to perform high-level tasks effectively. This understanding can be achieved through instance-aware semantic mapping, which involves reconstructing elements at the level of individual instances. Neural networks, the de facto solution for scene understanding, still face limitations such as overconfident incorrect predictions with out-of-distribution objects or generating inaccurate masks.Placing excessive reliance on these predictions makes the reconstruction susceptible to errors, reducing the robustness of the resulting maps and hampering robot operation. In this work, we propose Voxeland, a probabilistic framework for incrementally building instance-aware semantic maps. Inspired by the Theory of Evidence, Voxeland treats neural network predictions as subjective opinions regarding map instances at both geometric and semantic levels. These opinions are aggregated over time to form evidences, which are formalized through a probabilistic model. This enables us to quantify uncertainty in the reconstruction process, facilitating the identification of map areas requiring improvement (e.g. reobservation or reclassification). As one strategy to exploit this, we incorporate a Large Vision-Language Model (LVLM) to perform semantic level disambiguation for instances with high uncertainty. Results from the standard benchmarking on the publicly available SceneNN dataset demonstrate that Voxeland outperforms state-of-the-art methods, highlighting the benefits of incorporating and leveraging both instance- and semantic-level uncertainties to enhance reconstruction robustness. This is further validated through qualitative experiments conducted on the real-world ScanNet dataset.