Efficient exploration strategies are vital in tasks such as search-and-rescue missions and disaster surveying. Unmanned Aerial Vehicles (UAVs) have become particularly popular in such applications, promising to cover large areas at high speeds. Moreover, with the increasing maturity of onboard UAV perception, research focus has been shifting toward higher-level reasoning for multi-robot missions. However, autonomous navigation and exploration of previously unknown large spaces still constitute an open challenge, especially when the environment is cluttered and exhibits large and frequent occlusions due to high obstacle density, as is the case of forests. Moreover, the problem of long-distance wireless communication in such scenes can become a limiting factor, especially when automating the navigation of a UAV fleet. In this spirit, this work proposes an exploration strategy that enables multiple UAVs to quickly explore complex scenes in a decentralized fashion. By providing the decision-making capabilities to each UAV to switch between different execution modes, the proposed strategy is shown to strike a great balance between cautious exploration of yet completely unknown regions and more aggressive exploration of smaller areas of unknown space. This results in full coverage of forest areas in multi-UAV setups up to $\mathbf {30{\%}}$ faster than the state of the art.

On the pursuit of autonomous flying robots, the scientific community has been developing onboard real-time algorithms for localisation, mapping and planning. Despite recent progress, the available solutions still lack accuracy and robustness in many aspects. While mapping for autonomous cars had a substantive boost using deep-learning techniques to enhance LIDAR measurements using image-based depth completion, the large viewpoint variations experienced by aerial vehicles are still posing major challenges for learning-based mapping approaches. In this paper, we propose a depth completion and uncertainty estimation approach that better handles the challenges of aerial platforms, such as large viewpoint and depth variations, and limited computing resources. The core of our method is a novel compact network that performs both depth completion and confidence estimation using an image-guided approach. Real-time performance onboard a GPU suitable for small flying robots is achieved by sharing deep features between both tasks. Experiments demonstrate that our network outperforms the state-of-the-art in depth completion and uncertainty estimation for single-view methods on mobile GPUs. We further present a new photorealistic aerial depth completion dataset that exhibits more challenging depth completion scenarios than the established indoor and car driving datasets. The dataset includes an open-source, visual-inertial UAV simulator for photo-realistic data generation. Our results show that our network trained on this dataset can be directly deployed on real-world outdoor aerial public datasets without fine-tuning or style transfer.

With a large body of literature dedicated to ego-motion estimation and perception of a robot's workspace, the Robotics community has seen some impressive advances in self-localization and mapping, however, we are still far from general applicability of such approaches in real scenarios. Driven by the need for portable and low-cost solutions to relative pose estimation between Unmanned Aerial Vehicles (UAVs), in this work we propose a new framework to track a master UAV in real-time, carrying a known constellation of LED markers, from a slave UAV without any other pose estimation capability. This setup is especially interesting to aerial manipulation and close-up inspection of structures with low or no texture. Our approach is able to fuse the estimated master's pose with the slave's onboard inertial readings, supporting intermittent communication between the UAVs. Evaluation on both simulation and real indoor and outdoor experiments reveals that the proposed approach achieves unprecedented robustness to noise and occlusion, accuracy and speed of computation. All the code to reproduce this work is publicly available.

As vision-based navigation of small aircraft has been demonstrated to reach relative maturity, research into effective path-planning algorithms to complete the loop of autonomous navigation has been booming. Although the literature has seen some impressive works in this area, efficient path-planning that can be used in tasks such as inspection and coverage is still an open problem. In this spirit, this letter presents an online path-planning algorithm for fast exploration and 3D reconstruction of a previously unknown area of interest. Micro Aerial Vehicles (MAVs) are an ideal candidate for this task due to their maneuverability, but their limited computational power and endurance require efficient planning strategies. Popular sampling-based methods randomly sample the MAV's configuration space and evaluate viewpoints according to their expected information gain. Most often, however, valuable resources are spent on information gain calculations of unpromising viewpoints. This letter proposes a novel informed sampling approach that leverages surface frontiers to sample viewpoints only where high information gain is expected, leading to faster exploration. We study the impact of informed sampling in a wide range of photo-realistic scenes, and we show that our approach outperforms state-of-the-art exploration path planners in terms of both speed and reconstruction quality.

Robots operating in real-world settings often need to plan interactions with surrounding scene elements and therefore, it is crucial for them to understand their workspace at the level of individual objects. In this spirit, this work presents a novel approach to progressively build instance-level, dense 3D maps from color and depth cues acquired by either a moving RGB-D sensor or a camera-LiDAR setup, whose pose is being tracked. The proposed framework processes each input RGB image with a semantic instance segmentation neural network and uses depth information to extract a set of per-frame, semantically labeled 3D instance segments, which then get matched to object instances already identified in previous views. Following integration of these newly detected instance segments in a global volumetric map, an efficient label diffusion scheme that considers multi-view instance predictions together with the reconstructed scene geometry is used to refine 3D segmentation boundaries. Experiments on indoor benchmarking RGB-D sequences show that the proposed system achieves state-of-the-art performance in terms of 3D segmentation accuracy, while reducing the computational processing cost required at each frame. Furthermore, the applicability of the system to challenging domains outside the traditional office scenes is demonstrated by testing it on a robotic excavator equipped with a calibrated camera-LiDAR setup, with the goal of segmenting individual boulders in a highly cluttered construction scenario.

The efficiency of path-planning in robot navigation is crucial in tasks such as search-and-rescue and disaster surveying, but this is emphasized even more when considering multi-rotor aerial robots due to the limited battery and flight time. In this spirit, this work proposes an efficient, hierarchical planner to achieve comprehensive visual coverage of large-scale outdoor scenarios for small drones. Following an initial reconnaissance flight, a coarse map of the scene gets built in real-time. Then, regions of the map that were not appropriately observed are identified and grouped by a novel perception-aware clustering process that enables the generation of continuous trajectories ( sweeps ) to cover them efficiently. Thanks to this partitioning of the map into a set of tasks, we can generalize the planning to an arbitrary number of drones and perform a well-balanced workload distribution among them. We compare our approach against a state-of-the-art method for exploration and show the advantages of our pipeline in terms of efficiency for obtaining coverage in large environments.