This paper analyses the methods and technologies involved in flapping-wing flying robots (FWFRs), where the actuation of the flapping wing produces thrust and lift force that mimics birds’ and insects’ flight. The focus is on the evolution of the flapping-wing technology and the challenges in prototyping, modeling, navigation, and control. The mechanism for flapping production, frequency control of the flapping, and wing/tail control for positioning the robot are important topics for successful prototyping. The article includes the study of the dynamics and aerodynamics of the FWFR. Using the combination of flapping and gliding has led researchers to seek more energy savings through this hybrid-in-nature dynamic system, which benefits from the wind, a natural and free energy source. The paper reviews the dynamics, design, and categorization of flapping-wing systems; it also includes control and onboard intelligent functionalities, particularly environment perception for positioning and guidance, as well as obstacle detection and avoidance.

Autonomous flight in unknown, cluttered environments is still a major challenge in robotics. Existing obstacle avoidance algorithms typically adopt a fixed flight velocity, overlooking the crucial balance between safety and agility. We propose a reinforcement learning algorithm to learn an adaptive flight speed policy tailored to varying environment complexities, enhancing obstacle avoidance safety. A downside of learning-based obstacle avoidance algorithms is that the lack of a mapping module can lead to the drone getting stuck in complex scenarios. To address this, we introduce a novel training setup for the latent space that retains memory of previous depth map observations. The latent space is explicitly trained to predict both past and current depth maps. Our findings confirm that varying speed leads to a superior balance of success rate and agility in cluttered environments. Additionally, our memory-augmented latent representation outperforms the latent representation commonly used in reinforcement learning. Furthermore, an extensive comparison of our method with the existing state-of-the-art approaches Agile-autonomy and Ego-planner shows the superior performance of our approach, especially in highly cluttered environments. Finally, after minimal fine-tuning, we successfully deployed our network on a real drone for enhanced obstacle avoidance.

Lightweight aerial swarms have potential applications in scenarios where larger drones fail to operate efficiently. The primary foundation for lightweight aerial swarms is efficient relative localization, which enables cooperation and collision avoidance. Computing the real-time position is challenging due to extreme resource constraints. This letter presents an autonomous relative localization technique for lightweight aerial swarms without infrastructure by fusing ultra-wideband wireless distance measurements and the shared state information (e.g., velocity, yaw rate, height) from neighbors. This is the first fully autonomous, tiny, fast, and accurate relative localization scheme implemented on a team of 13 lightweight (33 grams) and resource-constrained (168 MHz MCU with 192 KB memory) aerial vehicles. The proposed resource-constrained swarm ranging protocol is scalable, and a surprising theoretical result is discovered: the unobservability poses no issues because the state drift leads to control actions that make the state observable again. By experiment, less than 0.2 m position error is achieved at the frequency of 16 Hz for as many as 13 drones. The code is open-sourced, and the proposed technique is relevant not only for tiny drones but can be readily applied to many other resource-restricted robots.

Sim-to-real transfer is a fundamental challenge in robot learning. Discrepancies between simulation and reality can significantly impair policy performance, especially if it receives high-dimensional inputs such as dense depth estimates from vision. We propose a novel depth transfer method based on domain adaptation to bridge the visual gap between simulated and real-world depth data. A Variational Autoencoder (VAE) is first trained to encode ground-truth depth images from simulation into a latent space, which serves as input to a reinforcement learning (RL) policy. During deployment, the encoder is refined to align stereo depth images with this latent space, enabling direct policy transfer without fine-tuning. We apply our method to the task of autonomous drone navigation through cluttered environments. Experiments in IsaacGym show that our method nearly doubles the obstacle avoidance success rate when switching from ground-truth to stereo depth input. Furthermore, we demonstrate successful transfer to the photo-realistic simulator AvoidBench using only IsaacGym-generated stereo data, achieving superior performance compared to state-of-the-art baselines. Real-world evaluations in both indoor and outdoor environments confirm the effectiveness of our approach, enabling robust and generalizable depth-based navigation across diverse domains.

The real-world application of small drones is mostly hampered by energy limitations. Neuromorphic computing promises extremely energy-efficient AI for autonomous flight but is still challenging to train and deploy on real robots. To reap the maximal benefits from neuromorphic computing, it is necessary to perform all autonomy functions end-to-end on a single neuromorphic chip, from low-level attitude control to high-level navigation. This research presents the first neuromorphic control system using a spiking neural network (SNN) to effectively map a drone's raw sensory input directly to motor commands. We apply this method to low-level attitude estimation and control for a quadrotor, deploying the SNN on a tiny Crazyflie. We propose a modular SNN, separately training and then merging estimation and control sub-networks. The SNN is trained with imitation learning, using a flight dataset of sensory-motor pairs. Post-training, the network is deployed on the Crazyflie, issuing control commands from sensor inputs at 500Hz. Furthermore, for the training procedure we augmented training data by flying a controller with additional excitation and time-shifting the target data to enhance the predictive capabilities of the SNN. On the real drone, the perception-to-control SNN tracks attitude commands with an average error of 3.0 degrees, compared to 2.7 degrees for the regular flight stack. We also show the benefits of the proposed learning modifications for reducing the average tracking error and reducing oscillations. Our work shows the feasibility of performing neuromorphic end-to-end control, laying the basis for highly energy-efficient and low-latency neuromorphic autopilots.

In high-speed quadcopter racing, finding a single controller that works well across different platforms remains challenging. This work presents the first neural network controller for drone racing that generalizes across physically distinct quadcopters. We demonstrate that a single network, trained with domain randomization, can robustly control various types of quadcopters. The network relies solely on the current state to directly compute motor commands. The effectiveness of this generalized controller is validated through real-world tests on two substantially different crafts (3-inch and 5-inch race quadcopters). We further compare the performance of this generalized controller with controllers specifically trained for the 3-inch and 5-inch drone, using their identified model parameters with varying levels of domain randomization (0%, 10%, 20%, 30%). While the generalized controller shows slightly slower speeds compared to the fine-tuned models, it excels in adaptability across different platforms. Our results show that no randomization fails sim-to-real transfer while increasing randomization improves robustness but reduces speed. Despite this trade-off, our findings highlight the potential of domain randomization for generalizing controllers, paving the way for universal AI controllers that can adapt to any platform.

Ego-Motion estimation is vital for drones when flying in GPS-denied environments. Vision-Based methods struggle when flight speed increases and close-by objects lead to difficult visual conditions with considerable motion blur and large occlusions. To tackle this, vision is typically complemented by state estimation filters that combine a drone model with inertial measurements. However, these drone models are currently learned in a supervised manner with ground-truth data from external motion capture systems, limiting scalability to different environments and drones. In this work, we propose a self-supervised learning scheme to train a neural-network-based drone model using only onboard monocular video and flight controller data (IMU and motor feedback). We achieve this by first training a self-supervised relative pose estimation model, which then serves as a teacher for the drone model. To allow this to work at high speed close to obstacles, we propose an improved occlusion handling method for training self-supervised pose estimation models. Due to this method, the root mean squared error of resulting odometry estimates is reduced by an average of 15%. Moreover, the student neural drone model can be successfully obtained from the onboard data. It even becomes more accurate at higher speeds compared to its teacher, the self-supervised vision-based model. We demonstrate the value of the neural drone model by integrating it into a traditional filter-based VIO system (ROVIO), resulting in superior odometry accuracy on aggressive 3D racing trajectories near obstacles. Self-Supervised learning of ego-motion estimation represents a significant step toward bridging the gap between flying in controlled, expensive lab environments and real-world drone applications. The fusion of vision and drone models will enable higher-speed flight and improve state estimation, on any drone in any environment.

Developing optimal controllers for aggressive high-speed quadcopter flight poses significant challenges in robotics. Recent trends in the field involve utilizing neural network controllers trained through supervised or reinforcement learning. However, the sim-to-real transfer introduces a reality gap, requiring the use of robust inner loop controllers during real flights, which limits the network's control authority and flight performance. In this paper, we investigate for the first time, an end-to-end neural network controller, addressing the reality gap issue without being restricted by an inner-loop controller. The networks, referred to as G&CNets, are trained to learn an energy-optimal policy mapping the quadcopter's state to rpm commands using an optimal trajectory dataset. In hover-to-hover flights, we identified the unmodeled moments as a significant contributor to the reality gap. To mitigate this, we propose an adaptive control strategy that works by learning from optimal trajectories of a system affected by constant external pitch, roll and yaw moments. In real test flights, this model mismatch is estimated onboard and fed to the network to obtain the optimal rpm command. We demonstrate the effectiveness of our method by performing energy-optimal hover-to-hover flights with and without moment feedback. Finally, we compare the adaptive controller to a state-of-the-art differential-flatness-based controller in a consecutive waypoint flight and demonstrate the advantages of our method in terms of energy optimality and robustness.

ROVIO is one of the state-of-the-art monocular visual inertial odometry algorithms. It uses an Iterative Extended Kalman Filter (IEKF) to align visual features and update the vehicle state simultaneously by including the feature locations in the state vector of the IEKF. This algorithm is single-core intensive, which allows the other cores to be used for other algorithms, such as object detection and path optimization. However, the computational cost of the algorithm grows rapidly with the maximum number of features to track. Each feature adds three new states (a 2D bearing vector and inverse depth), leading to bigger matrix multiplications that are computationally expensive. The main computational load of ROVIO is the iterative update step of the IEKF. In this work, we reduce the average computational cost of ROVIO by 40% on an NVIDIA Jetson TX2, without affecting the accuracy of the algorithm. This computational gain is mainly achieved by exploiting the sparse matrices in ROVIO. Furthermore, we reduce the computational peaks by pre-selecting new features based on their already calculated FAST score. The combination of both modifications allows us to run ROVIO on the computationally restricted Raspberry Pi Zero 2W.

Navigation is an essential capability for autonomous robots. In particular, visual navigation has been a major research topic in robotics because cameras are lightweight, power-efficient sensors that provide rich information on the environment. However, the main challenge of visual navigation is that it requires substantial computational power and memory for visual processing and storage of the results. As of yet, this has precluded its use on small, extremely resource-constrained robots such as lightweight drones. Inspired by the parsimony of natural intelligence, we propose an insect-inspired approach toward visual navigation that is specifically aimed at extremely resource-restricted robots. It is a route-following approach in which a robot's outbound trajectory is stored as a collection of highly compressed panoramic images together with their spatial relationships as measured with odometry. During the inbound journey, the robot uses a combination of odometry and visual homing to return to the stored locations, with visual homing preventing the buildup of odometric drift. A main advancement of the proposed strategy is that the number of stored compressed images is minimized by spacing them apart as far as the accuracy of odometry allows. To demonstrate the suitability for small systems, we implemented the strategy on a tiny 56-gram drone. The drone could successfully follow routes up to 100 meters with a trajectory representation that consumed less than 20 bytes per meter. The presented method forms a substantial step toward the autonomous visual navigation of tiny robots, facilitating their more widespread application.

Aggressive time-optimal control of quadcopters poses a significant challenge in the field of robotics. The state-of-the-art approach leverages reinforcement learning (RL) to train optimal neural policies. However, a critical hurdle is the sim-to-real gap, often addressed by employing a robust inner loop controller-an abstraction that, in theory, constrains the optimality of the trained controller, necessitating margins to counter potential disturbances. In contrast, our novel approach introduces high-speed quadcopter control using end-to-end RL (E2E) that gives direct motor commands. To bridge the reality gap, we incorporate a learned residual model and an adaptive method that can compensate for modeling errors in thrust and moments. We compare our E2E approach against a state-of-the-art network that commands thrust and body rates to an INDI inner loop controller, both in simulated and real-world flight. E2E showcases a significant 1.39-second advantage in simulation and a 0.17-second edge in real-world testing, highlighting end-to-end reinforcement learning's potential. The performance drop observed from simulation to reality shows potential for further improvement, including refining strategies to address the reality gap or exploring offline reinforcement learning with real flight data.

This Review discusses the main results obtained in training end-to-end neural architectures for guidance and control of interplanetary transfers, planetary landings, and close-proximity operations, highlighting the successful learning of optimality principles by the underlying neural models. Spacecraft and drones aimed at exploring our solar system are designed to operate in conditions where the smart use of onboard resources is vital to the success or failure of the mission. Sensorimotor actions are thus often derived from high-level, quantifiable, optimality principles assigned to each task, using consolidated tools in optimal control theory. The planned actions are derived on the ground and transferred on board, where controllers have the task of tracking the uploaded guidance profile. Here, we review recent trends based on the use of end-to-end networks, called guidance and control networks (G&CNets), which allow spacecraft to depart from such an architecture and to embrace the onboard computation of optimal actions. In this way, the sensor information is transformed in real time into optimal plans, thus increasing mission autonomy and robustness. We then analyze drone racing as an ideal gym environment to test these architectures on real robotic platforms and thus increase confidence in their use in future space exploration missions. Drone racing not only shares with spacecraft missions both limited onboard computational capabilities and similar control structures induced from the optimality principle sought but also entails different levels of uncertainties and unmodeled effects and a very different dynamical timescale.

Neuromorphic processing promises high energy efficiency and rapid response rates, making it an ideal candidate for achieving autonomous flight of resource-constrained robots. It can be especially beneficial for complex neural networks as are used for high-level visual perception. However, fully neuromorphic solutions also need to tackle low-level control tasks. Remarkably, it is currently still challenging to replicate even basic low-level controllers such as proportional-integral-derivative (PID) controllers. Specifically, it is difficult to incorporate the integral and derivative parts. To address this problem, we propose a neuromorphic controller that incorporates proportional, integral, and derivative pathways during learning. Our approach includes a novel input threshold adaptation mechanism for the integral pathway. This Input-Weighted Threshold Adaptation (IWTA) introduces an additional weight per synaptic connection, which is used to adapt the threshold of the post-synaptic neuron. We tackle the derivative term by employing neurons with different time constants. We first analyze the performance and limits of the proposed mechanisms and then put our controller to the test by implementing it on a microcontroller connected to the open-source tiny Crazyflie quadrotor, replacing the innermost rate controller. We demonstrate the stability of our bio-inspired algorithm with flights in the presence of disturbances. The current work represents a substantial step towards controlling highly dynamic systems with neuromorphic algorithms, thus advancing neuromorphic processing and robotics. In addition, integration is an important part of any temporal task, so the proposed Input-Weighted Threshold Adaptation (IWTA) mechanism may have implications well beyond control tasks.

Event cameras have recently gained significant traction since they open up new avenues for low-latency and low-power solutions to complex computer vision problems. To unlock these solutions, it is necessary to develop algorithms that can leverage the unique nature of event data. However, the current state-of-the-art is still highly influenced by the frame-based literature, and usually fails to deliver on these promises. In this work, we take this into consideration and propose a novel self-supervised learning pipeline for the sequential estimation of event-based optical flow that allows for the scaling of the models to high inference frequencies. At its core, we have a continuously-running stateful neural model that is trained using a novel formulation of contrast maximization that makes it robust to nonlinearities and varying statistics in the input events. Results across multiple datasets confirm the effectiveness of our method, which establishes a new state of the art in terms of accuracy for approaches trained or optimized without ground truth.