Research Projects

This work proposes a novel computationally efficient nonlinear model predictive controller (NMPC) for learning-based models. The proposed NMPC scheme uses a hybrid model of the dynamic system, including a nominal derived model and a learning-based model that compensates for the incomplete knowledge of the system, i.e., unmodeled dynamics. The NMPC is designed with a tailored cost function that takes into account the learned-dynamics of the system. The cost function is formulated without stabilizing terminal conditions required for stabilization. Moreover, the proposed scheme facilitates the computation of the shortest possible stabilizing prediction horizon that guarantees the asymptotic stability of the closed-loop system. The proposed scheme is applied to an unmanned aerial vehicle (UAV) for validation. The performance of the proposed scheme is investigated through extensive numerical simulations and compared against the state-of-the-art traditional NMPC and traditional learning-based NMPC schemes proposed in literature. The results show superior trajectory tracking performance of the proposed learning-based NMPC scheme at short prediction horizons.

Related Publications: C8

This work proposes a new filtering-based depth enhanced visual internal navigation system (DE-VINS) with external disturbance observation. This filter resolves the drifting and degraded performance of drag force model VINS filters at hovering conditions and during the existence of external disturbances. A theoretical nonlinear observability analysis is performed to verify the filter design. The performance of the proposed DE-VINS is investigated through two sets of numerical simulations using a Matlab simulator and compared against the state-of-the-art drag force VINS filters. The results show improved performance of the DE-VINS in terms of estimation accuracy and consistency at zero-velocity flight (hovering) during the existence of external disturbances while estimating the magnitude and direction of the disturbance force. The proposed DE-VINS filter has observability guarantees of estimating the magnitude and direction of the applied disturbance even at zero velocity while maintaining consistent performance. Future work will include the design of the observability-constrained DE-VINS to improve the consistency of the filter and experimental validation of the proposed filter using our newly developed multi-sensor platform.

Related Publications: C6, C7

During the development of this new multi-sensor platform, point cloud coloring was performed by matching each 2D camera pixel to its corresponding 3D point cloud.

This work presents the design of a stability-guaranteed nonlinear model predictive controller for quadrotor-type micro aerial vehicles to operate robustly on fast trajectories. The basic controller structure operates without having to use terminal conditions in the optimization problem. As a result, the controller is computationally less demanding and provides more stable closed-loop performance than traditional nonlinear predictive control schemes. The work presents a detailed stability analysis without terminal costs or terminal constraints and proves the asymptotic stability and necessary conditions for recursive feasibility of the system. The work derives the growth bound sequence that enables obtaining the shortest possible prediction horizon for stability. The proposed analysis provides the necessary conditions to implement the controller while using the shortest stabilizing prediction horizon as compared to major traditional predictive control schemes reported in the literature. This particular feature enables the proposed controller to perform fast optimization and hence the capability to implement fast trajectories using feedback regularization. In order to demonstrate the validity of this new proposed control scheme, first, several MATLAB simulations are conducted to demonstrate the improved performance of the controller especially when the quadrotor vehicle follows fast trajectories. Real-time lab experiments are also conducted to validate the performance of the proposed scheme for point stabilization (hovering) and trajectory tracking problems. The results show that the proposed scheme can stabilize the system with a relatively short prediction horizon, with a fast convergence rate, and a small tracking error.

Related Publications: J4, J5

This work presents the design of an interacting multiple model (IMM) filter for visual-inertial navigation (VIN) of MAVs. VIN of MAVs in practice typically uses a single system model for its state estimator design. However, MAVs can operate in different scenarios such as aggressive flights, hovering flights, and under high external disturbance requiring changing constraints imposed on the estimator model. This study proposes the use of a conventional VIN and a drag force VIN in an error-state IMM filtering framework to address the need for multiple models in the estimator. The work uses an epipolar geometry constraint for the design of the measurement model for both filters to realize computationally efficient state updates. Observability of the proposed modifications to VIN filters (drag force model, and epipolar measurement model) are analyzed, and observability based consistency rules are derived for the two filters of the IMM. Monte Carlo numerical simulations validate the performance of the observability constrained IMM, which improved the accuracy and consistency of the VINS for changing flight conditions and external wind disturbance scenarios. Experimental validation is performed using the EuRoC dataset to evaluate the performance of the proposed IMM filter design. The results show that the IMM outperforms stand-alone filters used in the IMM filtering bank by switching between the filters based on the residual likelihood of the models. 

Related Publications: J3, C5

In this work, we address the problem of real-time control and localization of an autonomous art exhibit designed using Mecanum wheeled omnidirectional mobile robots. Currently, the exhibit is animated using open-loop control as there are no practical means of implementing a localization system in art galleries. This paper proposes a nonlinear model predictive controller (NMPC) supported by a Decawave localization system for trajectory tracking of the robots. A Robot Operating System (ROS) based NMPC is implemented to control the motion of the robot for a smooth and drift-free trajectory tracking. The Automatic Control and Dynamic Optimization (ACADO) toolkit is used to find the optimal control action while considering all constraints on system states and inputs. A Local Positioning System (LPS) is implemented using Decawave 1001 modules to provide position feedback with an Optitrack motion capture system providing ground truth information for validation. The proposed control approach is first evaluated using the Virtual Robot Experimentation Platform (V-REP), which additionally provides means of trajectory design and simulation of the robot in the desired exhibition space. Thereafter, laboratory experiments are conducted to evaluate the performance of the proposed control and localization systems. The results show a smooth and drift-free performance of the system with less than 10\% error of the robot size, which can be deployed in gallery spaces with minimal setup requirements.

Related Publications: C4 

In general, Directional Steering System (DSS) has been established for well drilling in the oilfield in order to accomplish high reservoir productivity and to improve accessibility of oil reservoirs in complex locations. DSS facilitates the accessibility of the oil reservoirs if the reservoir is consisting of wide surface zone in a thin horizontal layer.

My research work was divided into two main parts:

1- Control and Optimization of Rotary Steerable System (RSS) which is the current state of the art in directional drilling. In this work we address the problem of real-time control of autonomous RSS with unknown formation friction and rock strength. The work presents an online control scheme for real-time optimization of drilling parameters to maximize rate of penetration and minimize the deviation from the planned well bore trajectory.  Nonlinear model for the drilling operation was developed using energy balance equation.  The model parameters have been adaptively estimated at each control iteration using Particle Swam Optimization (PSO) technique to tackle any disturbances or variations in the formation properties. 

Related Publications: J2, C1, C2

2- Control and Optimization of autonomous Quad-Rotor Directional Drilling which has 4-DC motors. A novel feedback linearization controller to cancel the nonlinear dynamics of a the DSS is proposed. The proposed controller design problem is formulated as an optimization problem for optimal settings of the controller feedback gains. Gravitational Search Algorithm (GSA) is developed to search for optimal settings of the proposed controller. The objective function considered is to minimize the tracking error and drilling efforts.

Related Publications: J1