Precise system identification is an important aspect of adequate control design and parameter definition to allow for accurate and reliable navigation. While this is well known in robotics, the community working with small rotorcraft Unmanned Aerial Vehicles (UAVs) has yet to discover the benefits. In contrast to existing work, which often performs offline or deterministic (i.e. closed-form) system identification, we present a probabilistic approach to the online estimation of system identification parameters and self-calibration states. Instead of decoupling system identification and state estimation for vehicle control, we merge the entire process into a holistic probabilistic framework to allow self-awareness and self-healing. Our observability analysis shows that most of the system identification parameters are observable and converge quickly to the optimal value using a combination of inertial cues, dynamic modeling, and an additional exteroceptive sensor. We support our theoretical findings with extensive tests simulating realistic data in Gazebo.

The research community presented significant advances in many different Visual-Inertial Navigation System (VINS) algorithms to localize mobile robots or hand-held devices in a 3D environment. While authors of the algorithms of-ten do compare to, at that time, existing competing approaches, their comparison methods, rigor, depth, and repeatability at later points in time have a large spread. Further, with existing simulators and photo-realistic frameworks, the user is not able to easily test the sensitivity of the algorithm under examination with respect to specific environmental conditions and sensor specifications. Rather, tests often include unwillingly many polluting effects falsifying the analysis and interpretations. In addition, edge cases and corresponding failure modes often remain undiscovered due to the limited breadth of the test sequences. Our unified evaluation framework allows, in a fully automated fashion, a reproducible analysis of different VINS methods with respect to specific environmental and sensor parameters. The analyses per parameter are done over a multitude of test sets to obtain both statistically valid results and an average over other, potentially polluting effects with respect to the one parameter under test to mitigate biased interpretations. The automated performance results per method over all tested parameters are then summarized in unified radar charts for a fair comparison across authors and institutions.

State of the art visual-inertial odometry approaches suffer from the requirement of high gradients and sufficient visual texture. Even direct photometric approaches select a subset of the image with high-gradient areas and ignore smooth gradients or generally low-textured areas. In this work, we show that taking all image information (i.e. every single pixel) enables visual-inertial odometry even on areas with very low texture and smooth gradients, inherently interpolating and estimating the scene with no texture based on its informative surrounding. This information propagation is only possible as we estimate all states and their uncertainties (robot pose, extrinsic sensor calibration, and scene depth) jointly in a fully dense filter framework. Our complexity reduction approach enables real-time execution despite the large size of the state vector. Compared to our previous basic feasibility study on this topic, this work includes higher order covariance propagation and improved state handling for a significant performance gain, thorough comparisons to state-of-the-art algorithms, larger mapping components with uncertainty, self-calibration capability, and real-data tests.

We propose a fully dense direct filter-based visual-inertial odometry method estimating both pixel depth for all pixels and robot state simultaneously, having all uncertainties in the same state vector. Due to the fully dense method, our approach works even in low-textured areas with very low, smooth gradients (i.e. scenes where feature based or semi-dense approaches fail). Our algorithm performs in real-time on a CPU with a time complexity linearly dependent on the amount of pixels in the provided image. To achieve this, we propose complexity reduction methods for fast matrix inversion, exploiting specific structures of the covariance matrix. We provide both simulated and real-world results in low-textured areas with a smooth gradient.

Many of today's Visual-Inertial Odometry (VIO)frameworks work well under regular motion but have issues and need special treatment under special motion. Here, special does not imply bad or corrupted data but stands for increased difficulty to treat clean data. Common special motion for VIO are large feature displacement due to fast motion close to a scene and zero motion phases not providing sufficient baseline. In this paper we present a feature and frame selection approach which seamlessly handles all motion scenarios without the need of (error prone)motion case identification and subsequent case-specific heuristics. We further show that this approach allows to eliminate features in the state vector (persistent features)altogether while still being able to inherently handle zero motion phases. This reduces computational complexity while maintaining the ability to hover in place. We integrate our frame selection approach into our own VIO algorithm and compare its performance against three state-of-the-art algorithms with real data on a real platform. While our approach shows slightly higher global drift it is the only algorithm that can reliably estimate the pose over a large motion spectrum from fast scale change down to zero motion.