Two controller performances are assessed for generalization in the path following task of autonomously backing up a tractor-trailer. Starting from random locations and orientations, paths are generated to loading docks with arbitrary pose using Dubins Curves. The combination vehicles can be varied in wheelbase, hitch length, weight distributions, and tire cornering stiffness. The closed form calculation of the gains for the Linear Quadratic Regulator (LQR) rely heavily on having an accurate model of the plant. However, real-world applications cannot expect to have an updated model for each new trailer. Finding alternative robust controllers when the trailer model is changed was the motivation of this research. Reinforcement learning, with neural networks as their function approximators, can allow for generalized control from its learned experience that is characterized by a scalar reward value. The Linear Quadratic Regulator and the Deep Deterministic Policy Gradient (DDPG) are compared for robust control when the trailer is changed. This investigation quantifies the capabilities and limitations of both controllers in simulation using a kinematic model. The controllers are evaluated for generalization by altering the kinematic model trailer wheelbase, hitch length, and velocity from the nominal case. In order to close the gap from simulation and reality, the control methods are also assessed with sensor noise and various controller frequencies. The root mean squared and maximum errors from the path are used as metrics, including the number of times the controllers cause the vehicle to jackknife or reach the goal. Considering the runs where the LQR did not cause the trailer to jackknife, the LQR tended to have slightly better precision. DDPG, however, controlled the trailer successfully on the paths where the LQR jackknifed. Reinforcement learning was found to sacrifice a short term reward, such as precision, to maximize the future expected reward like reaching the loading dock. The reinforcement learning agent learned a policy that imposed nonlinear constraints such that it never jackknifed, even when it wasn't the trailer it trained on.

Unmanned aerial vehicles (UAV's) have recently gained popularity in military, civil service, agriculture, commercial, and hobby use. This is due in part to their affordability, which comes from advances in component technology. That technology includes microelectromechanical systems (MEMS) for inertial sensing, microprocessor technology for sequential algorithm processing, field programmable gate arrays (FPGA's) for parallel data processing, camera technology, global navigation satellite systems (GNSS's) for navigation, and battery technology such as the high energy density of lithium polymer batteries. Despite the success of the technology to date, there remains development before UAV's should be flying alongside manned aircraft or over populated areas. One concern is that UAV electronics are not as safe, reliable or robust as manned-aircraft electronics because UAV's are not certified by the FAA. Another concern for UAV operation is with control algorithms and sensors, particularly in the estimation of the aircraft state, which is the position, velocity, and orientation of the aircraft. Some problems, such as numerical stability of a control algorithm or flight in windy and turbulent conditions have only been solved for certain conditions of wind, weather, or maneuvers. Outside those conditions, the actual orientation of a flying craft can mislead to the control system, and the control system may not be able to recover without a crash. When pilots fly manned aircraft in instrument meteorological conditions, or conditions of limited visibility of the ground, terrain, and obstacles, the pilot must fly in a manner which avoids abrupt maneuvers which could disturb accuracy of the aircraft's instruments. In a UAV without a pilot, there is a need to estimate the position and orientation of a UAV in an absolute manner unambiguous relative to the Earth. The position and orientation estimate must not depend on carefully controlled flight paths, but instead the estimate must be robust in the presence of UAV flight dynamics. This thesis describes the design, implementation, and evaluation of a hardware platform for GPS based orientation sensing research. In this work, we considered a receiver with three or four RF sections, each connected to an antenna in a triangular or tetrahedral pyramid constellation. Specific requirements for the receiver hardware and functionality were created. Circuitry was designed to meet the requirements using commercial off-the-shelf (COTS) radio frequency (RF) modules, a mid-sized microcontroller, an FPGA, and other supporting components. A printed circuit board (PCB) was designed, fabricated, assembled, and tested. A GPS baseband processor was designed and coded in Verilog hardware description language. The design was synthesized and loaded to the FPGA, and the microcontroller was programmed to track satellites. With the hardware platform implemented, live satellite signals were found and tracked, and experiments were performed to explore the validity of GPS based orientation sensing using short antenna baselines. The platform successfully allows the user to develop correlator designs and explore carrier phase based orientation measurement using only software/Verilog modifications. Initial results of carrier phase based orientation sensing are promising, but the presence of multipath signal interference shows room for improvement to the baseband processing code.

An investigation was conducted into the feasibility of using an unconventional flight technique, asymmetric flight, to improve overall efficiency of solar aircraft. In this study, asymmetric flight is defined as steady level flight in a non-wings-level state in- tended to improve solar incidence angle. By manipulating aircraft orientation through roll angle, solar energy collection is improved but aerodynamic efficiency is worsened due to the introduction of additional trim drag. A point performance model was devel- oped to investigate the trade-off between improvement in solar energy collection and additional drag associated with asymmetric flight. A mission model with a focus on aircraft orbits was then developed via integration of the point performance model over a set of discrete points. It is shown that there is a non-zero bank angle where optimal net power is achieved for a given aircraft orientation, flight condition, and sun position. The study also shows that there is improvement in overall efficiency over conventional flight for various orbit shapes and winds aloft. This indicates that there is potential value in not only flight path planning, but also in orientation planning for solar aircraft.

Unmanned aerial systems (UAS) carry no pilot on board, yet they still require live operators to handle critical functions such as mission planning and execution. Humans also interpret the sensor information provided by these platforms. This applies to all classes of unmanned aerial vehicles (UAV's), including the smaller portable systems used for gathering real-time reconnaissance during military operations in urban terrain. The need to quickly and reliably train soldiers to control small UAS operations demands that the human-system interface be intuitive and easy to master. In this study, participants completed a series of tests of spatial ability and were then trained (in simulation) to teleoperate a micro-unmanned aerial vehicle equipped with forward and downward fixed cameras. Three aspects of the human-system interface were manipulated to assess the effects on manual control mastery and target detection. One factor was the input device. Participants used either a mouse or a specially programmed game controller (similar to that used with the Sony Playstation 2 video game console). A second factor was the nature of the flight control displays as either continuous or discrete (analog v. digital). The third factor involved the presentation of sensor imagery. The display could either provide streaming video from one camera at a time, or present the imagery from both cameras simultaneously in separate windows. The primary dependent variables included: 1) time to complete assigned missions, 2) number of collisions, 3) number of targets detected, and 4) operator workload. In general, operator performance was better with the game controller than with the mouse, but significant improvement in time to complete occurred over repeated trials regardless of the device used. Time to complete missions was significantly faster with the game controller, and operators also detected more targets without any significant differences in workload compared to mouse users. Workload on repeated trials decreased with practice, and spatial ability was a significant covariate of workload. Lower spatial ability associated with higher workload scores. In addition, demographic data including computer usage and video gaming experience were collected and analyzed, and correlated with performance. Higher video gaming experience was also associated with lower workload.

The CalPoly Spacecraft Attitude Dynamics Simulator mimics the rotational dynamics of a spacecraft in orbit and acts as a testbed for spacecraft attitude control system development and demonstration. Prior to this thesis, the simulator platform and several subsystems had been designed and manufactured, but the total simulator system was not yet capable of closed-loop attitude control. Previous attempts to make the system controllable were primarily mired by data transport performance. Rather than exporting data to an external command computer, the strategy implemented in this thesis relies on a compact computer onboard the simulator platform to handle both attitude control processing and data acquisition responsibilities. Software drivers were created to interface the computer’s data acquisition boards with Matlab, and a Simulink library was developed to handle hardware interface functions and simplify the composition of attitude control schemes. To improve the usability of the system, a variety of actuator control, hardware testing, and data visualization utilities were also created. A closedloop attitude control strategy was adapted to facilitate future sensor installations, and was tested in numerical simulation. The control model was then updated to interface with the simulator hardware, and for the first time in the project history, attitude control was performed onboard the CalPoly spacecraft attitude dynamics simulator. The demonstration served to validate the numerical model and to verify the functionality of the entire simulator system.

Due to mass and volume constraints on planetary missions, the development of control techniques that do not require fuel are of big interest. For those planets that have a dense enough atmosphere, aerodynamic drag can play an important role. The use of atmospheric differential drag for formation keeping was first proposed by Carolina L. Leonard in 1986, and has been proven to work in Earth atmosphere by many missions. Moreover, atmospheric drag has been used in the Mars atmosphere as aerobraking technique to decelerate landing vehicles, and to circularize the orbit of the spacecraft. Still, no literature was available related to formation flying on Mars. To analyze the use of differential drag on the Mars atmosphere, the researcher accessed the two high resolution models available: NASA’s Mars-GRAM and ESA’s Mars Climate Database. These models allowed the simulation of conditions that a spacecraft would experience while in orbit around the planet. To explore the feasibility, the researcher first conducted a study where Mars atmosphere density was compared to Earth atmosphere, determining its applicability. Then, a simulation using MATLAB® was conducted, using a Keplerian two-body problem including the effects of Mars zonal harmonics (i.e. J2) and drag perturbations. Two 6U CubeSat were used in the simulation with deployable drag plates of different sizes, giving the possibility of having five differential drag scenarios as means of formation control. The conclusions showed that, although with some limitations, the use of differential drag as means of autonomous formation flying and proximity operations control is feasible using proven techniques previously validated in Low Earth Orbit. Lyapunov control was selected as the control strategy, where three different methods were evaluated and compared.

This thesis contains the design and simulation test results for the implementation of a new all-wheel drive (AWD) torque vectoring (TV) control system. A separate algorithm using standard control methods is included in this study for a comparison. The proposed controller was designed to be applied to an AWD independent drive electric vehicle, however the main concepts can be re-purposed for other vehicle drive train configurations. The purpose of the control system is to assist the driver in achieving a desired vehicle trajectory whilst also maintaining stability and control of the vehicle. This is accomplished by measuring various real time parameters of the vehicle and using this information as feedback for the control system to act on. The focus of this thesis resides on the controller. Hence, this study assumes perfect observation of feedback parameters, therefore some uncertainties are not accounted for. Using feedback parameters, the control system will manage wheel slip whilst simultaneously generating a torque around the center of gravity of the vehicle by applying a torque differential between the left and right wheels. The proposed TV algorithm is simulated in MATLAB/Simulink along with another separate TV algorithm for comparison. Both algorithms are comprised of two main parts: a slip ratio controller applied to each wheel individually and stability controller that manages yaw rate and side slip of the vehicle. The new algorithm leverages the super twisting algorithm for the slip ratio controller and uses a fusion of a linear quadratic regulator with the integral term of a super twisting algorithm to implement the yaw rate and side slip controller. The other algorithm used for comparison derives its implementation for the slip ratio controller and yaw rate and side slip controllers from simple and standard first order sliding mode control methods. Both control algorithms were tested in three different main tests: anti-lock braking, sine dwell (SD) steering, and constant steering angle (CSA) tests. To increase the comprehensive nature of the study, the SD and CSA tests were simulated at 3 speeds (30,50, and 80 mph) and the steering angle parameter was varied from 2 to 24 degrees in increments of 2. The result of this study proves that the proposed controller is a feasible option for use in theory. Simulated results show advantages and disadvantages of the new controller with respect to the standard comparison controller. Both controllers are also shown to provide positive impacts on the vehicle response under most test conditions.

Using a new implementation of a NASA flight simulation of the Quiet Short-Haul Research Aircraft, autopilots were designed to be capable of flying both straight in (ILS) approaches, and circling (SNI) approaches. A standard glideslope coupler was sufficient for most conditions, but a standard Proportional-Integral-Derivative (PID) based localizer tracker was not sufficient for maintaining a lateral track on the SNI course. To track the SNI course, a feed-forward system, using GPS steering provided much better results. NASA and the FAA embrace the concept of a Simultaneous, Non-Interfering (SNI) approach as a way to increase airport throughput while reducing the noise footprints of aircraft on approach. The NASA concept for the SNI approach for Short Takeoff and Landing (STOL) aircraft involves a straight in segment flown above the flight path of a normal approach, followed by a spiraling descent to the runway. As this is a procedure that would be utilized by regional airliners, it is assumed that it would be conducted under Instrument Flight Rules (IFR). GPS or INS guidance would be required to fly this approach, and it is likely that it would be necessary to fly the approach with a coupled autopilot: a stabilized, curving, instrument approach to decision altitude would be exceedingly difficult to fly. The autopilots in many current commuter and general aviation aircraft, however, were designed before the event of GPS, and do not have provisions for tracking curved paths. This study identifies problem areas in implementing the SNI circling approach on aircraft and avionics as they stand today and also gives examples of what can be done for the SNI approach to be successful.

The paper describes the background and concepts behind a master’s thesis platform known as COMET (Constrained Optimization of Multiple-dimensions for Efficient Trajectories) created for mission designers to determine and evaluate suitable interplanetary trajectories. This includes an examination of the improvements to the global optimization algorithm, Differential Evolution, through a cascading search space pruning method and decomposition of optimization parameters. Results are compared to those produced by the European Space Agency’s Advanced Concept Team’s Multiple Gravity Assist Program. It was found that while discrepancies in the calculation of ΔV’s for flyby maneuvers exist between the two programs, COMET showed a noticeable improvement in its ability to avoid premature convergence and find highly isolated solutions.

On-board gimbal systems for camera stabilization in helicopters are typically based on linear models. Such models, however, are inaccurate due to system nonlinearities and complexities. As an alternative approach, artificial neural networks can provide a more accurate model of the gimbal system based on their non-linear mapping and generalization capabilities. This thesis investigates the applications of artificial neural networks to model the inertial characteristics (on the azimuth axis) of the inner gimbal in a gyro-stabilized multi-gimbal system. The neural network is trained with time-domain data obtained from gyro rate sensors of an actual camera system. The network performance is evaluated and compared with measured data and a traditional linear model. Computer simulation results show the neural network model fits well with the measured data and significantly outperforms a traditional model.