Thomas Bewley

Eulerian nonlinear uncertainty propagation methods often suffer from finite domain limitations and computational inefficiencies. A recent approach to this class of algorithm, Grid-based Bayesian Estimation Exploiting Sparsity, addresses the first challenge by dynamically allocating a discretized grid in regions of phase space where probability is non-negligible. However, the design of the original algorithm causes the second challenge to persist in high-dimensional systems. This paper presents an architectural optimization of the algorithm for CPU implementation, followed by its adaptation to the CUDA framework for single GPU execution. The algorithm is validated for accuracy and convergence, with performance evaluated across distinct GPUs. Tests include propagating a three-dimensional probability distribution subject to the Lorenz'63 model and a six-dimensional probability distribution subject to the Lorenz'96 model. The results imply that the improvements made result in a speedup of over 1000 times compared to the original implementation.

Eulerian nonlinear uncertainty propagation methods often suffer from finite domain limitations and computational inefficiencies. A recent approach to this class of algorithm, Grid-based Bayesian Estimation Exploiting Sparsity, addresses the first challenge by dynamically allocating a discretized grid in regions of phase space where probability is non-negligible. However, the design of the original algorithm causes the second challenge to persist in high-dimensional systems. This paper presents an architectural optimization of the algorithm for CPU implementation, followed by its adaptation to the CUDA framework for single GPU execution. The algorithm is validated for accuracy and convergence, with performance evaluated across distinct GPUs. Tests include propagating a three-dimensional probability distribution subject to the Lorenz '63 model and a six-dimensional probability distribution subject to the Lorenz '96 model. The results imply that the improvements made result in a speedup of over 1000 times compared to the original implementation.

Moment-based estimation filters have successfully aided spacecraft navigation for decades. However, future missions plan to venture into deep-space regimes with significant round-trip light-time telecommunication delays, operate in unstable, quasi-periodic orbits, and perform highly precise, low-altitude flybys of outer planet moons. These complex trajectories may necessitate ensemble-based filters for accurate estimation over realistic measurement cadences. To mitigate the inherent risk associated with testing novel navigation software, ensemble filters must be accurate, efficient, and robust. Grid-based, Bayesian Estimation Exploiting Sparsity, a high-dimensional Godunov-type finite volume method that propagates the full probability distribution function, demonstrates strong overall performance across all these criteria when compared with the contemporary landscape of filters. These qualities are exhibited via a Bayesian investigation in which the state uncertainty of a Saturn-Enceladus Distant Prograde Orbit is propagated, incorporating infrequent, nonlinear measurement updates. Along with root mean square error, we use the Bhattacharyya coefficient, a non-normal metric for measuring the dissimilarity between distributions, and the Effective Sampling Size, a measure of particle degeneracy, to quantitatively ascertain that in this application, Grid-based, Bayesian Estimation Exploiting Sparsity outperforms the other ensemble filters assessed, though it comes at a nontrivial computational cost.

A Series Elastic Actuator (SEA) is a mechanical system that consists of a motor, an elastic material, and an end-effector placed in a series. By actuating the motor and loading the elastic material, the SEA can generate forces at the end-effector. Once the elastic material is loaded, the SEA can continuously produce forces without requiring additional motor actuation. This paper proposes a torsion spring-motor integrated SEA, specifically designed for applying consistent forces on a robot arm that can potentially contribute to achieving stability in robot poses. The SEA configuration comprises a motor securely attached to the body frame and a torsion spring connected between the robot arm and the motor shaft. When the robot arm contacts the ground, the actuating motor causes a deflection in the torsion spring. A force sensor positioned beneath the robot arm measures the pushing forces resulting from the deflection of the torsion spring. The torques can then be calculated by multiplying the distance between the SEA shaft and the robot arm by the measured force. Other design variations are also discussed with different assembly orders. The hardware tests show that the difference between the measured and estimated torques computed using Hooke's law of two suggested designs was an average of 0.0847 N-m and the normalized root mean square error of 13.48%.

The current landscape of orbital uncertainty propagation methods inadequately addresses the state estimation problem for nonlinear systems. In relatively low-perturbed regimes, or when measurement updates are frequent, state estimation methods that assume Gaussian uncertainty are valid, and errors resulting from linearizing the dynamics about an estimate are negligible. However, as novel space mission design techniques begin to exploit the chaoticity of N-body dynamics to efficiently explore new regimes of space, the Gaussianity assumption is often violated, and linearization errors accumulate. Uncertainty propagation methods that do not assume Gaussianity or linearize about an estimate are computationally expensive. Moreover, both classes of methods often disregard epistemic uncertainty, or the uncertainty of the model. To address the current limitations of orbital uncertainty propagation, we introduce a higher-dimensional extension to an existing Bayesian estimation algorithm that efficiently propagates the probability distribution function of a state governed by nonlinear dynamics. By adjusting the computational architecture of the algorithm and considering the dynamics of the system, we scale the existing, three-dimensional technique with poor time complexity to an efficient, four-dimensional one. The result is a robust, second-order accurate, time-adaptive, explicit time-marching scheme with the capability of propagating uncertainty governed by chaotic, nonlinear dynamics.

In this work, we propose a lightweight dynamic object filtering algorithm for building LiDAR-based static point cloud maps in realtime. On one hand, we propose an egocentric motion detection method of using improved ICP to register 3D clusters and extract their poses and twists to identify dynamic objects. One the other hand, we connect the proposed dynamic object filter with LiDAR-based SLAM algorithms to build point cloud maps and validate the effectiveness of the proposed methodology on both our custom dataset and SemanticKITTI. We also compare the performance of the proposed method against state-of-the-art methods in terms of both filtering accuracy and processing time. As experimentally verified on SemanticKITTI, our method yields promising performance with relatively small time costs and therefore has great potential to be used as point cloud data source for a number of LiDAR-inertial-visual fusion mapping methods.

The cable-driven Boat Motion Simulator is a mechanical system designed to replicate the intricate 6 degrees of freedom motions experienced by a boat. This simulator consists of 8 cables and a moving platform. The platform is connected to the cables, and its position is controlled by actuating motors that adjust the cable lengths accordingly. To accurately replicate the desired boat motions, optimizing the design parameters that offer sufficient workspace for the simulator is crucial. This paper focuses on analyzing the workspace using three design parameters: height, width, and cable attachment points. Examining the workspace under various parameter configurations aims to optimize the design and ensure that the simulator provides a sufficient range of motion to replicate complex boat motions accurately. To assess the workspace, we developed a simulation model and an algorithm incorporating three methods: static equilibrium analysis, cable-to-cable interactions, and cable-to-platform interference detection algorithm. This research provides insights into achieving an optimized design for the Cable-driven Boat Motion Simulator, enabling realistic boat motion replication.

This paper considers the 3D rotation estimation of a moving platform from 2D images captured by a camera. Assume that a circular pattern marker is on the flight deck of a ship and quadrotor hovers on the center of a platform. The quadrotor has a camera that captures 2D images of markers and measures the distance between markers. The circular pattern of markers changes to an ellipse as the platform rotates. The ellipse equation can be derived from marker positions on an ellipse, and a platform's rotation sequence can be determined by leveraging geometry and trigonometric functions. As the platform's rotation can be estimated, the quadrotor is able to decide the timing that platform rotation is within the quadrotor's capacity to land safely. A simulation and a hardware test were performed to verify the estimation method, and the estimation error was discussed.

Multirotors have become the most popular UAV or aerial robot category due to their structural simplicity and ability to take off and land vertically. However, most multirotors suffer from short airborne time and range due to limited battery capacity. Thus we propose a novel hybrid multirotor design, QuadGlider, to increase its capability of traveling long distances with minimal battery consumption. QuadGlider is inspired by the body structure and gliding mechanism of gliding animals. The airframe design of QuadGlider imitates flying squirrels' skeletal and muscular structure with implementation of servo motor to actuate its compliant membrane wings. Therefore, it can transition from quadrotor flight mode to forward gliding mode via morphing and lowering motor speeds to save power. In this work, we first present the conceptual design of QuadGlider. Next, we model its flight dynamics in different gliding scenarios. Then, the design is verified in computational fluid dynamics simulations of gliding scenarios with angles of attack of <tex>$0^{\mathrm{o}}-30^{\mathrm{o}}$</tex>. At last, preliminary gliding experiments are conducted at low Reynolds numbers of around <tex>$Re=4_{*}7_{*}\times 10^{5}$</tex> The equilibrium glide simulation gives a maximum glloe rauo of 4.27: 1 at a takeoff velocity of 14.0 m/s, while the experimental result indicates a glide ratio of 2.97: 1 at a takeoff velocity of 4.17 m/s.

Examples and implementations of various robotic mechanisms, devices, components, systems and techniques are provided, including multimodal robotic devices and systems. For example, a multimodal robot can be configured to autonomously reconfigure between two or more primary modes of operation. Such robots may be used in a wide range of applications, including reconnaissance, exploration, search and rescue, military, sports, personal assistance, education, and entertainment and toys. Described examples of multimodal robots can be wheeled robots that use two or more drive wheels to perform various motions and operations.