We present a new learning-based method for identifying safe and navigable regions in off-road terrains and unstructured environments from RGB images. Our approach consists of classifying groups of terrains based on their navigability levels using coarse-grained semantic segmentation. We propose a bottleneck transformer-based deep neural network architecture that uses a novel group-wise attention mechanism to distinguish between navigability levels of different terrains. Our group-wise attention heads enable the network to explicitly focus on the different groups and improve the accuracy. We show through extensive evaluations on the RUGD and RELLIS-3D datasets that our learning algorithm improves visual perception accuracy in off-road terrains for navigation. We compare our approach with prior work on these datasets and achieve an improvement over the state-of-the-art mIoU by 6.74-39.1% on RUGD and 3.82-10.64% on RELLIS-3D. In addition, we deploy our method on a Clearpath Jackal robot. Our approach improves the performance of the navigation algorithm in terms of average progress towards the goal by 54.73% and the false positives in terms of forbidden region by 29.96%.
Decentralized deployment of drone swarms usually relies on inter-agent communication or visual markers that are mounted on the vehicles to simplify their mutual detection. This letter proposes a vision-based detection and tracking algorithm that enables groups of drones to navigate without communication or visual markers. We employ a convolutional neural network to detect and localize nearby agents onboard the quadcopters in real-time. Rather than manually labeling a dataset, we automatically annotate images to train the neural network using background subtraction by systematically flying a quadcopter in front of a static camera. We use a multi-agent state tracker to estimate the relative positions and velocities of nearby agents, which are subsequently fed to a flocking algorithm for high-level control. The drones are equipped with multiple cameras to provide omnidirectional visual inputs. The camera setup ensures the safety of the flock by avoiding blind spots regardless of the agent configuration. We evaluate the approach with a group of three real quadcopters that are controlled using the proposed vision-based flocking algorithm. The results show that the drones can safely navigate in an outdoor environment despite substantial background clutter and difficult lighting conditions. The source code, image dataset, and trained detection model are available at https://github.com/lis-epfl/vswarm.
The use of drones for aerial cinematography has revolutionized several applications and industries that require live and dynamic camera viewpoints such as entertainment, sports, and security. However, safely controlling a drone while filming a moving target usually requires multiple expert human operators; hence the need for an autonomous cinematographer. Current approaches have severe real-life limitations such as requiring fully scripted scenes, high-precision motion-capture systems or GPS tags to localize targets, and prior maps of the environment to avoid obstacles and plan for occlusion. In this work, we overcome such limitations and propose a complete system for aerial cinematography that combines: (1) a vision-based algorithm for target localization; (2) a real-time incremental 3D signed-distance map algorithm for occlusion and safety computation; and (3) a real-time camera motion planner that optimizes smoothness, collisions, occlusions and artistic guidelines. We evaluate robustness and real-time performance in series of field experiments and simulations by tracking dynamic targets moving through unknown, unstructured environments. Finally, we verify that despite removing previous limitations, our system achieves state-of-the-art performance. Videos of the system in action can be seen at https://youtu.be/ZE9MnCVmumc
We present a method for detecting and mapping trees in noisy stereo camera point clouds, using a learned 3-D object detector. Inspired by recent advancements in 3-D object detection using a pseudo-lidar representation for stereo data, we train a PointRCNN detector to recognize trees in forest-like environments. We generate detector training data with a novel automatic labeling process that clusters a fused global point cloud. This process annotates large stereo point cloud training data sets with minimal user supervision, and unlike previous pseudo-lidar detection pipelines, requires no 3-D ground truth from other sensors such as lidar. Our mapping system additionally uses a Kalman filter to associate detections and consistently estimate the positions and sizes of trees. We collect a data set for tree detection consisting of 8680 stereo point clouds, and validate our method on an outdoors test sequence. Our results demonstrate robust tree recognition in noisy stereo data at ranges of up to 7 meters, on 720p resolution images from a Stereolabs ZED 2 camera. Code and data are available at https://github.com/brian-h-wang/pseudolidar-tree-detection.
To achieve high-levels of autonomy, modern robots require the ability to detect and recover from anomalies and failures with minimal human supervision. Multi-modal sensor signals could provide more information for such anomaly detection tasks; however, the fusion of high-dimensional and heterogeneous sensor modalities remains a challenging problem. We propose a deep learning neural network: supervised variational autoencoder (SVAE), for failure identification in unstructured and uncertain environments. Our model leverages the representational power of VAE to extract robust features from high-dimensional inputs for supervised learning tasks. The training objective unifies the generative model and the discriminative model, thus making the learning a one-stage procedure. Our experiments on real field robot data demonstrate superior failure identification performance than baseline methods, and that our model learns interpretable representations. Videos of our results are available on our website: https://sites.google.com/illinois.edu/supervised-vae .
We present Terrain Traversability Mapping (TTM), a real-time mapping approach for terrain traversability estimation and path planning for autonomous excavators in an unstructured environment. We propose an efficient learning-based geometric method to extract terrain features from RGB images and 3D pointclouds and incorporate them into a global map for planning and navigation for autonomous excavation. Our method used the physical characteristics of the excavator, including maximum climbing degree and other machine specifications, to determine the traversable area. Our method can adapt to changing environments and update the terrain information in real-time. Moreover, we prepare a novel dataset, Autonomous Excavator Terrain (AET) dataset, consisting of RGB images from construction sites with seven categories according to navigability. We integrate our mapping approach with planning and control modules in an autonomous excavator navigation system, which outperforms previous method by 49.3% in terms of success rate based on existing planning schemes. With our mapping the excavator can navigate through unstructured environments consisting of deep pits, steep hills, rock piles, and other complex terrain features.