The INSANE data sets are a versatile collection of Micro Aerial Vehicle (MAV) data sets designed to support research in autonomous robotic platforms. Accurate and robust localization is a key requirement for these platforms to navigate safely in various dynamic environments. The data sets provide different scenarios with multiple levels of difficulty for localization methods, including indoor and outdoor environments and challenging Mars analog scenarios. The extensive sensor suite includes various sensor categories, such as multiple Inertial Measurement Units (IMUs) and cameras, and each data set provides highly accurate ground truth. The data sets and post-processing tools are available for download and aim to reflect real-world scenarios and sensor effects to support research on machine learning-based sensor signal enhancement methods for improved localization.

This letter presents a new generic formulation for a gyroscope aided attitude estimator using N direction measurements. The approach incorporates navigation, extrinsic calibration for all direction sensors, and gyroscope bias states in a single geometric structure. The proposed filter-based estimator improves the transient response, and the asymptotic bias and extrinsic calibration estimation compared to state-of-the-art approaches. The estimator is verified in simulations and tested in real-world experiments.

This research paper presents the CNS Flight Stack, a high-level flight stack for unmanned aerial vehicles (UAVs) designed to enable safe and fully autonomous long-duration missions. The framework is platform-agnostic, uses low compute complexity, and can be extended with other sensor modalities, integrity checks, and mission modules. The CNS Flight Stack has been tested in over 450 real-world flights, making it a valuable tool for the UAV community, and it is freely available for external mission planners and localization modules.

This research paper presents a modular sensor-fusion framework that allows for the addition and removal of sensors during runtime in dynamic environments. The framework handles system and sensor initialization, measurement updates, and switching of asynchronous multi-rate sensor information with sensor self-calibration. It also has the ability to handle delayed measurements, out-of-sequence updates, and monitor sensor health. The introduced true-modularity is based on covariance segmentation, allowing the processing of propagation and updates on a per-sensor basis. The framework was tested in a precision landing scenario using GNSS, barometer, and vision measurements in both simulation and real-world scenarios. The framework is open-sourced and available for the community to use.

In the context of the AMADEE-18 analog Mars mission, a team of researchers focused on testing the localization of an unmanned aerial vehicle (UAV) on Mars. Due to the computational limitations of robot helicopters, accurate localization is crucial for autonomous navigation and path planning. In the absence of a global positioning system, the team tested a visual-inertial odometry (VIO) algorithm and camera in a Mars-like analog environment to evaluate its feasibility for localization on Mars. Flight datasets from various terrains were used to challenge the functionality of the VIO algorithm and provide insights into the desired surface structure, texture, and mission times for surface relative navigation of UAVs on Mars. The study provides valuable information for the challenges associated with human-based exploration of the Red Planet.

Introducing our fully autonomous rotorcraft unmanned aerial system (UAS) - the solution to long-term observation missions in remote locations without any human intervention. With limited flight times of traditional UAS and the need for humans in the loop, our platform offers a solution for uninterrupted operations over days or weeks. Our technology addresses two critical areas: platform autonomy and vision-based precision landing for automated energy replenishment. Our high-level autonomous decision-making hierarchy and vision-based landing system uses AprilTag fiducials to estimate landing pad pose for accuracy. We have conducted indoor and outdoor experiments, where the UAS executed 16, 48, and 22 flights with a flying-to-charging ratio of 1 to 10. With no human interaction required, this is the first quadrotor system to operate long-term outdoors.

This study introduces equivariant filtering as a powerful tool for refining state estimation in unmanned aerial vehicles (UAVs). Equivariant filters offer faster convergence rates and greater robustness to initial state errors compared to traditional methods by exploiting mathematical symmetries. The research demonstrates their effectiveness in sensor fusion using GNSS and IMU data and validates their practicality in real-world scenarios with the INSANE Dataset. This innovation holds immense promise for enhancing UAV state estimation in real-world environments.

This paper highlights the importance of extracting meaningful information from sensory data in mobile robotics. It introduces a method that combines AI-based object pose estimation from images with inertial measurement unit (IMU) data. This fusion enables accurate multi-object relative state estimation in a 6-DoF context, demonstrated through real-world experiments. The approach’s self-calibrating capabilities ensure reliable and reproducible results.

This paper presents a Radar-Inertial Odometry (RIO) method that leverages advanced techniques from vision research to accurately estimate a robot’s position and velocity using radar data. The approach combines past robot poses, radar measurements, and Inertial Measurement Unit (IMU) readings in an Extended Kalman Filter (EKF) framework. This method is particularly valuable for Unmanned Aerial Vehicles (UAVs) operating in challenging environments without GNSS, and demonstrates its effectiveness in real flight experiments.

This article discusses the challenges of localization filters for underwater vehicles and presents the Modular and Robust Sensor-Fusion Framework (MaRS) as a solution. MaRS is extended to work with underwater vehicles and their environment and allows efficient use of asynchronous sensors, handles measurement outliers and outages, and includes sensor-frame initialization and online extrinsic calibration methods. Tests using a small remotely operated vehicle (ROV) show improved handling of sensors and state estimation results in real harbor-like environments.

This article describes a data-driven approach to autonomously control hydraulic manipulators with minimal system information. The hydraulic actuation dynamics are modeled using actuator networks, which emulates the real system in a simulation environment. The approach uses a neural network control policy based on end-effector position tracking learned through Reinforcement Learning (RL) with Ornstein-Uhlenbeck process noise for efficient exploration. The proposed approach is implemented on a hydraulic forwarder crane to track the desired position of the end-effector in 3D space, and the results demonstrate the feasibility of deploying the learned controller directly on the real system.

This article discusses the challenges of estimating the end-effector pose for Suspended Cable-Driven Parallel Robots (SCDPR), due to the curved shape of cables known as cable sag. The proposed real-time pose estimation algorithm considers cable sag to reduce estimation errors and also employs an Inertial Measurement Unit (IMU) to improve accuracy for dynamic motion. The approach reduces Root Mean Squared Error (RMSE) compared to standard methods and is evaluated on a real system with ground truth pose information.

This paper proposes a probabilistic approach for online system identification and self-calibration in small rotorcraft Unmanned Aerial Vehicles (UAVs) for improved control design and navigation. The approach integrates the system identification and state estimation processes into a single framework, allowing for self-awareness and self-healing, and uses a combination of inertial cues, dynamic modeling, and an additional sensor for convergence to the optimal value. The results are supported by simulations using realistic data in Gazebo.

This paper introduces a new approach to improve state propagation for unmanned aerial vehicles using AI algorithms to preprocess high-speed inertial data. Two network architectures are evaluated, an LSTM-based approach and a Transformer encoder architecture, with the former outperforming the latter. The networks are designed to directly accept input data at variable rates and provide sufficient temporal history for good performance while maintaining a high propagation rate of preprocessed IMU samples. Results show significant improvements in propagation error even for long IMU propagation times.

In this paper, we propose a method to address the performance loss or failure of state estimation and navigation frameworks caused by local magnetic distortions. The use of magnetometers on-board mobile platforms suffer from these issues if magnetic distortions are not detected and mitigated. The paper shows the importance of representing magnetic variation in a spherical coordinate system instead of Cartesian coordinates, which can improve estimator consistency and enable accurate and fast mitigation of magnetic disturbances. The approach is validated by performing tests with simulated and real-world data on embedded hardware. The proposed method has the potential to lead to better system state estimates in magnetically distorted areas.

This paper introduces a new approach to filtering data from sensors on multiple autonomous agents, such as robots or drones, to estimate the agents’ positions and orientations relative to each other. The proposed method, called Quaternion-based Error-State Extended Kalman Filter (Q-ESEKF), uses advanced mathematical techniques to make the filtering more accurate and efficient. It also incorporates Collaborative State Estimation (CSE), which allows each agent to estimate its own state independently while taking into account measurements from other agents. The paper presents simulation results demonstrating the effectiveness of the approach on two benchmark datasets in 3D.

This research proposes a fully autonomous small rotorcraft UAS for long-term observation missions that requires no human intervention. The system addresses two key technologies critical for such a system: full platform autonomy and vision-based precision landing for automated energy replenishment. Experimental results show up to 11 hours of fully autonomous operation in indoor and outdoor environments, demonstrating the system’s capability. This technology could revolutionize unmanned aerial vehicle surveillance and monitoring applications, enabling observations at precise locations over long periods of time without the need for human intervention.

The extended abstract presents the AMAZE helicopter as part of the robotic team involved in the exploration cascade of the Mars Analog Simulation AMADEE-20. Analog research is used to prepare and train for future Mars exploration missions, with the Austrian Space Forum (OeWF) having conducted 12 Mars analog missions as part of the AMADEE research program. The AMADEE program develops strategies to detect life on extraterrestrial planets and ensures high fidelity for all OeWF analog missions. The AMAZE helicopter will be used for surface exploration during the simulation in the Ramon Crater, Negev Desert, Israel, which aims to replicate conditions on Mars.