GAIA: Ground-Aerial maps Integration for increased Autonomy outdoors

由于多个机器人智能体协作完成同一任务的灵活性，多机器人地图在针对诸如现场检查、人工搜救和果园监控等各种问题时总是成为更有需求的应用。在这种观点下，机器人需要相互通信，以共享关于它们的相对位置和周围环境的信息。综合这些信息，可以全面了解环境和特派团的进展情况。共享信息的一个更简单的例子是由单个机器人在其操作过程中构建的本地地图。在文献中，多地图集成通常被描述为地图合并问题，其中地图被表示为占用网格并通过寻找重叠部分来拼接在一起，或者被描述为多机器人同时定位和映射(SLAM)问题，其中单个机器人的位姿图被构建并使用图论进行连接。然而，这仍然是一个有趣和开放的研究问题，尽管在过去的20年里，人们在机器人定位方面做出了努力。与单一机器人使用案例不同，多机器人地图集成需要更高级别的抽象来识别多个地图中共有的元素，从而使集成高效且实时。特别是，在这个我们称之为“增强户外自主性的地面-航空地图集成”(GAIA)的项目中，我们的目标是拥有一支具有互补行为、运动和感知能力的不同机器人车队的场景，这使得地图集成问题比同类车队(例如，只使用地面机器人)更加困难。为了解决这一问题，本项目致力于利用人类对场景的理解来集成多视角机器人地图，特别是我们计划利用机器人观察中的语义信息来更好地理解场景，识别场景中存在的实体，并利用这些信息将多视角观察整合到单一地图表示中。更具体地说，在盖亚和这次电话会议的范围内，我们解决了农业领域的多机器人地图集成问题，在农业领域，机器人解决方案可以代表一种改变游戏规则的技术。在田间部署自主代理以协助(如果不是取代)人类工人监测和收获任务的可能性开启了一场专注于精准农业和可持续发展的新革命。事实上，机器人配备了专用的软件和硬件，可以通过收集降雨量、土壤水分和土壤成分的数据来帮助农民，从而帮助他们进行更有针对性的干预。更具体地说，地面机器人提供了对农作物更近、更详细的检查视角，而无人机则允许在更短的时间内观察更大的田地。无人机提供主动传感功能，按需完成和更新部分完整的地图，同时提高其置信度(我们对地图的信任程度)，例如通过绘制位于田野中的人类工人或任何其他与农业有关的工具(例如拖拉机或手推车)。UGV的路径规划者可以利用这些更新的信息来提高地面平台的部署效率，避免那些受阻的路径。

Multi-robot mapping is becoming an always more demanded application for targeting a large variety of problems, such as site inspection, human search and rescue, and orchard monitoring, due to the flexibility of having multiple robotic agents acting cooperatively for completing the same task. In this view, the robots need to communicate one with another to share information on their relative position and their surrounding. Integrating this information allows for obtaining a global overview of the environment and the progress status of the mission. One of the simpler examples of shared information is represented by the local maps built by individual robots during their operations. In the literature, multiple maps integration is usually cast as a map merging problem, in which maps are represented as occupancy grids and stitched together by looking for overlapping parts, or as a multi-robot Simultaneous Localisation and Mapping (SLAM) problem, where graphs of individual robots' pose are built and connected using graph theory. However, this is still an interesting and open research problem despite the effort put into robot localisation in the last twenty years. Differently from the single robot use case, multi-robot maps integration needs a higher level of abstraction to identify which elements are common across multiple maps so to make the integration efficient and real-time. In particular, within this project we call "Ground-Aerial maps Integration for increased Autonomy Outdoors" (GAIA), we target the scenario of having a fleet of heterogeneous robots characterised by complementary behaviours, movements and perception capabilities, making the maps integration problem even harder than homogeneous fleet (e.g., using only ground robots). To solve this problem, this project focuses on exploiting the human understanding of a scene so as to integrate multi-perspective robotic maps.In particular, we plan to use the semantic information in robotic observations to have a better understanding of the scene, to identify which entities are present in it, and to leverage such information so as to integrate multi-perspective observations into a single map representation. More specifically, within GAIA and the scope of this call, we tackle the problem of multi-robot maps integration in the agricultural domain, where robotics solutions can represent a game-changer technology. The possibility of deploying autonomous agents in the field to assist, if not replace, human workers in monitoring and harvesting tasks opens up a new revolution focused on precision agriculture and sustainability. Indeed, robots are equipped with dedicated software and hardware that can assist farmers by collecting data on rainfall, soil moisture and soil composition, so to help them make more target interventions. More specifically, the ground robot offers a closer and more detailed inspection point of view over the crops, while the UAV allows observing a larger field in a shorter time. The UAV offers active sensing capabilities to complete and update a partially complete map on demand, while improving its level of confidence (how much we trust the map) by, for example, mapping human workers or any other agriculture-related tools (e.g., tractors or trolleys) located in the fields. This updated information can be exploited by the UGV's path planner to make the ground platform's deployment more efficient, avoiding those obstructed paths.