Learning-based visual ego-motion estimation is promising yet not ready for navigating agile mobile robots in the real world. In this article, we propose CUAHN-VIO, a robust and efficient monocular visual-inertial odometry (VIO) designed for micro aerial vehicles (MAVs) equipped with a downward-facing camera. The vision frontend is a content-and-uncertainty-aware homography network (CUAHN). Content awareness measures the robustness of the network toward non-homography image content, e.g. 3-dimensional objects lying on a planar surface. Uncertainty awareness refers that the network not only predicts the homography transformation but also estimates the prediction uncertainty. The training requires no ground truth that is often difficult to obtain. The network has good generalization that enables “plug-and-play” deployment in new environments without fine-tuning. A lightweight extended Kalman filter (EKF) serves as the VIO backend and utilizes the mean prediction and variance estimation from the network for visual measurement updates. CUAHN-VIO is evaluated on a high-speed public dataset and shows rivaling accuracy to state-of-the-art (SOTA) VIO approaches. Thanks to the robustness to motion blur, low network inference time (∼23 ms), and stable processing latency (∼26 ms), CUAHN-VIO successfully runs onboard an Nvidia Jetson TX2 embedded processor to navigate a fast autonomous MAV. ... Read more

Biological sensing and processing is asynchronous and sparse, leading to low-latency and energy-efficient perception and action. In robotics, neuromorphic hardware for event-based vision and spiking neural networks promises to exhibit similar characteristics. However, robotic implementations have been limited to basic tasks with low-dimensional sensory inputs and motor actions because of the restricted network size in current embedded neuromorphic processors and the difficulties of training spiking neural networks. Here, we present a fully neuromorphic vision-to- control pipeline for controlling a flying drone. Specifically, we trained a spiking neural network that accepts raw event-based camera data and outputs low-level control actions for performing autonomous vision-based flight. The vision part of the network, consisting of five layers and 28,800 neurons, maps incoming raw events to ego-motion estimates and was trained with self-supervised learning on real event data. The control part consists of a single decoding layer and was learned with an evolutionary algorithm in a drone simulator. Robotic experiments show a successful sim-to- real transfer of the fully learned neuromorphic pipeline. The drone could accurately control its ego-motion, allowing for hovering, landing, and maneuvering sideways—even while yawing at the same time. The neuromorphic pipeline runs on board on Intel’s Loihi neuromorphic processor with an execution frequency of 200 hertz, consuming 0.94 watt of idle power and a mere additional 7 to 12 milliwatts when running the network. These results illustrate the potential of neuromorphic sensing and processing for enabling insect-sized intelligent robots. ... Read more

When deployed onboard micro air vehicles (MAVs) with limited processing power, visual ego-motion estimation solutions face an efficiency-accuracy trade-off. This paper proposes an aerodynamic-model-aided approach that emphasizes time efficiency over estimation accuracy. A linear drag force model of propellers guarantees bounded estimation errors in the velocity components orthogonal to the shafts of propellers and the attitude relative to the gravity direction. Feature point correspondences are extracted from the monocular image stream to compute the relative heading angle and translational direction, which is fused with inertial measurements by an extended Kalman filter (EKF) in a loosely coupled manner. The proposed approach shows balanced performance in accuracy and efficiency. It also has robustness to situations where vision information becomes unavailable. ... Read more

In this paper, we propose an obstacle avoidance solution for a 34-gram quadcopter equipped with a monocular camera. The perception of obstacles is tackled by a lightweight convolutional neural network predicting a dense depth map from a captured grey-scale image. The depth network performs self-supervised learning and thus requires no ground-truth labels that are costly to acquire. Based on the depth map, the control strategy is implemented by a behavior state machine that balances the efficiency to explore the environment and the safety of avoiding obstacles. In real-world flight experiments, our solution demonstrates the efficacy of predicting trust-worthy depth maps and a stable control strategy in various cluttered environments. ... Read more

In this paper, we propose a learning-based lightweight visual-inertial odometry (VIO) based on an uncertainty-aware pose network and an extended Kalman filter (EKF). The pose network serving as the VIO vision front-end predicts the relative motion of the camera between consecutive image frames and estimates the prediction uncertainty. The training of the pose network can be conducted without requiring ground-truth labels. The distributions of visual measurements are fused with inertial measurements by an EKF that is the VIO back-end. Evaluations show that the proposed VIO fails to outperform a state-of-the-art feature-point-based VIO solution in accuracy. But it has high time efficiency, translational motion estimation with metric scale, estimation of gravity direction, and generalization to new environments. So, unlike most works on learning-based visual ego-motion estimation in the literature, the proposed VIO can be directly deployed on an MAV. The comparative studies of supervision signals and forms of translational motion prediction provide insights that can contribute to future research. ... Read more

Micro air vehicles (MAVs) have shown significant potential in modern society. The development in robotics and automation is changing the roles of MAVs from remotely controlled machines requiring human pilots to autonomous and intelligent robots. There is an increasing number of autonomous MAVs involved in outdoor operations. In contrast, the deployment of MAVs in GPS-denied environments is relatively less practiced. The speed when flying indoors is often slow. One reason is that MAVs are surrounded by obstacles. But it should also be noticed that ego-motion estimation becomes more difficult to remain reliable during faster flight. The reason for this is that fast motion brings challenges to the robustness and computational efficiency of ego-motion estimation solutions based on the limited onboard sensing and processing capacities. The challenge to robustness is that the motion blur induced by agile maneuvers reduces the amount of available visual information needed by the current mainstream ego-motion estimation solutions, given the fact that frame-based cameras are the primary sensor for most lightweight MAVs. The challenge of computational efficiency comes from the strong desire for smaller and smaller MAVs to better fit cluttered environments. Moreover, to compensate for the decrease in robustness, additional computational power is required to detect known landmarks or visual processing that better copes with motion blur. This dissertation responds to the challenges by investigating novel ego-motion estimation approaches that combine robustness and efficiency. First, the goal of higher efficiency in the context of traditional visual feature points is pursued, albeit at the cost of reduced accuracy. The targeted scenarios are where known landmarks exist, such as gates in autonomous drone racing. The proposed velocity estimator’s mission is to navigate the MAV until the next landmark appears in the field of view and corrects the accumulated drift in the position estimation. To prevent drift over time, a simple linear drag force model is used for estimating the pitch and roll angles of the MAV with respect to the gravity vector and its velocity within the horizontal plane of the propellers. The translational motion direction and the relative yaw angle are efficiently calculated from the correspondences of feature points using a RANSAC-based linear algorithm... ... Read more

In the field of visual ego-motion estimation for Micro Air Vehicles (MAVs), fast maneuvers stay challenging mainly because of the big visual disparity and motion blur. In the pursuit of higher robustness, we study convolutional neural networks (CNNs) that predict the relative pose between subsequent images from a fast-moving monocular camera facing a planar scene. Aided by the Inertial Measurement Unit (IMU), we mainly focus on translational motion. The networks we study have similar small model sizes (around 1.35MB) and high inference speeds (around 10 milliseconds on a mobile GPU). Images for training and testing have realistic motion blur. Departing from a network framework that iteratively warps the first image to match the second with cascaded network blocks, we study different network architectures and training strategies. Simulated datasets and a self-collected MAV flight dataset are used for evaluation. The proposed setup shows better accuracy over existing networks and traditional feature-point-based methods during fast maneuvers. Moreover, self-supervised learning outperforms supervised learning. Videos and open-sourced code are available at https://github. com/tudelft/PoseNet_Planar ... Read more