David M. Bevly

Trajectory duplication using relative position information for automated ground vehicle convoys

A strategy to enhance the accuracy of path following for autonomous ground vehicles in a convoy is presented. GPS carrier measurements are used to estimate relative position with sub-two centimeter accuracy and a change in position to millimeter accuracy. These estimates are used in conjunction with three methods presented that enable a following vehicle to replicate a lead vehiclepsilas path of travel while both are in motion and not in sight of one another. Accuracies of the methods achieved in simulation are shown with discussion on the benefits and shortcomings of each method. Simulation results show a 1.6 meter error at a 50 m following distance. Discussion explains the inaccuracies are due to the limitations inherent in the selected vehicle controller and not necessarily in the trajectory duplication methods.

Observer design for differentiable Lipschitz nonlinear systems with time-varying parameters

This paper develops observer design techniques in a unified framework for both time invariant and parameter varying Lipschitz nonlinear systems that are differentiable w.r.t. state variables. First, a new sufficient condition for asymptotic convergence is developed for both the extended Luenberger observer and a two-DOF nonlinear observer for time-invariant nonlinear systems. In addition to ensuring asymptotic convergence, extension of this observer design technique to optimization of a L2 performance criterion is presented, which enables the observer to handle the unknown disturbance inputs as well as ensure robustness to model uncertainty. Next, augmentation of this technique to parameter varying nonlinear (PVNL) systems is developed. Different from methods suggested in the LPV literature, a simple but non-conservative finite dimensional relaxation method for quadratic parameter dependent LMIs is presented. These results constitute perhaps the first systematic observer design methodology in literature for PVNL systems. Finally, a simulation result for vehicle slip angle estimation is presented to verify the performance of the developed observer design methods.

Implementation details of a deeply integrated GPS/INS software receiver

The goal of this paper is to describe the implementation details of an embedded GPS/INS software receiver. Several different methods of improving performance in difficult environments have been studied for the past few decades, but vector tracking and deep integration (or ultra-tight coupling) has been especially popular in the past few years. Vector tracking algorithms boast the ability to maintain signal lock in weak signal-to-noise ratio environments such as urban canyons or heavy foliage. Another touted benefit is the ability to instantly reacquire GPS signal lock after an outage. The addition of an inertial sensor to aid the vector tracking algorithms is known as deep integration or ultra-tight coupling. The addition of this inertial sensor further boosts immunity to jamming and receiver dynamics. Implementing vector tracking and deep integration on a real-time platform does not come without its drawbacks. Besides the typical real-time deadlines, a full traditional (non-vector) GPS receiver must be implemented to initialize the vector tracking algorithm. The addition of an IMU also requires the algorithm to track attitude and inertial bias states, which increases the size and complexity of the algorithm. In this paper, the design of an FPGA hardware platform to perform both traditional and vector tracking/deep integration is described in detail along with some preliminary computational results.

Robust observer design for Lipschitz nonlinear systems using quadratic polynomial constraints

This paper discusses the observer design for the uncertain Lipschitz nonlinear systems. A new stability analysis method for the Lure problem is first presented. Then, a nonlinear observer is proposed so that the observer error dynamic model can be transformed to an equivalent Lure system in which the input-output relationship of the nonlinear memoryless block is belong to the semi-algebraic set defined by several quadratic polynomial constraints. A sufficient condition for the exponential stability of the observer error dynamics is formulated in terms of the feasibility of linear matrix inequalities (LMIs).

Impact of carrier to noise power density, platform dynamics, and IMU quality on deeply integrated navigation

In this paper, the authors investigate how the Carrier to Noise power density ratio (C/N0), platform dynamics, and differing Inertial Measurement Unit (IMU) quality affect the performance of Deeply Integrated (DI) algorithms. Two different DI algorithms are described in detail and analyzed using a high fidelity GPS simulator. The first algorithm is a Vector Delay/Frequency Lock Loop (VDFLL). The second algorithm is a Deeply Integrated GPS/INS system with differing grades of IMUpsilas. The ability of the algorithms to operate at low C/N0 levels and in high dynamics is investigated empirically with the GPS simulator. The VDFLL algorithm can successfully track the received GPS signals through 2 g, 4 g, and 8 g coordinated turns at 19 dB-Hz. Initial results of the Deeply Integrated GPS/INS algorithm show its operation through the 2 g, 4 g, and 8 g coordinated turn at 16 dB-Hz with a tactical grade IMU.

Observer Design for Parameter Varying Differentiable Nonlinear Systems, With Application to Slip Angle Estimation

This technical note develops observer design techniques in a unified framework for both time invariant and parameter varying Lipschitz nonlinear systems that are differentiable w.r.t. state variables. First, a new sufficient condition for asymptotic convergence is developed for Arcaks two-DOF nonlinear observer for time-invariant nonlinear systems. In addition to ensuring asymptotic convergence, extension of this observer design technique to optimization of a

Performance comparison of single and dual frequency closely coupled GPS/INS relative positioning systems

L_{2}

Relating local vision measurements to global navigation satellite systems using waypoint based maps

performance criterion is presented, which improves disturbance rejection performance of the observer. Next, augmentation of this technique to parameter varying nonlinear (PVNL) systems is developed. Different from methods suggested in the LPV literature, a simple but non-conservative finite dimensional relaxation method for quadratic parameter dependent LMIs is presented. These results constitute perhaps the first systematic observer design methodology in literature for PVNL systems. Finally, the performance of the developed observers is evaluated for estimation of slip angle using the commercial vehicle dynamics software CARSIM. Unlike previous results which use a LTI or a time invariant nonlinear model for observer design, this technical note presents a PVNL observer guaranteed to work with continuously varying velocity of the vehicle.

Lane Tracking using Multilayer Laser Scanner to Enhance Vehicle Navigation and Safety Systems

Single and dual frequency closely coupled GPS/INS relative positioning systems have been developed for use in automated ground vehicle convoys. The accuracy of the GPS carrier phase measurement is exploited to produce a high precision relative position vector between vehicles. The inertial systems of the lead vehicle and following vehicle are aligned using closely coupled GPS/INS filters on each vehicle and are used to output a high rate solution.

Vehicle Lane Position Estimation with Camera Vision using Bounded Polynomial Interpolated Lines

Roughly 50% of all the traffic fatalities are due to lane departures [2]. There is great interest in advanced driver assistance systems that prevent unintended lane departure. Currently there are passive lane detection systems that warn the driver of unintended land departure know as land departure warning (LDW) systems which rely on cameras to track lane markings. LDW systems base solely off camera measurements are prone to failures due to poor environmental lighting or poor lane marking coverage. Combining the measurements from multiple sensors will create a much more robust LDW system that is not prone to failures.