Research Areas

Stochastic Control Theory

Traditionally, optimal control theory assumes that there are no probabilistic (more broadly stochastic) errors in the observation and state models, governed by differential equations, while stochastic optimal control generalizes optimal control theory to include these errors.  While having its roots in the calculus of variations (some 350+ years ago), most of the previous work to date has focused on control of stochastic linear systems.  This severely limits applicability to realistic systems whose dynamics are governed by nonlinear Ito diffusions.  Although standard linear techniques can sometimes be applied to nonlinear systems, there are no performance guarantees at all in practice. For the first time ever, a control law for a large class of nonlinear systems minimizing sample path deviations has been developed, providing the first-ever true expansion of stochastic optimal control for nonlinear systems. 

Autonomous Localization, Mapping, and Object Identification in Outdoor Lawns 

This project – funded by Samsung and the SC Department of Commerce – enhances existing technology by utilizing a Simultaneous Localization and Mapping (SLAM) algorithm in combination with a convolutional neural network to perform semantic SLAM. In contrast to metric SLAM, the problem of including semantic information is in its infancy. However, I see great potential to enhance a robot’s autonomy, facilitate tasks, and enable task-planning instead of path-planning through the association of semantic concepts to geometric objects in the environment. 

Framework for Accommodating Emerging Autonomous Vehicles

I am participating in a funded research project by the Center for Connected Multimodal Mobility (C2M2) motivated by the emergence of connected and autonomous vehicles and their potential for disruptively transforming freight transportation. The project aims at demonstrating that significant fuel savings and reduction in congestion can be achieved by routing a fleet of autonomous vehicles over a highway system such that platooning opportunities can be taken advantage of. Work on this project addresses a particular aspect of great interest to me: There will be a critical transition period of incremental steps in which the promises of autonomous technology will be tested on roadways and evaluated by both free market forces and the court of public opinion. It is research addressing this transition phase in mobility which is urgently needed.

Two-wheeled Mobile Manipulators under Uncertain Surface Conditions

Existing two-wheeled manipulators are very limited in coping with rapidly varying parameters and external forces and exhibit a very low tolerance to slip due to being statically unstable. Our efforts employ model-based estimation of key mobile base states enabled by fusing GPS data with automotive grade IMUs and other sensors along with error compensation. Subsequent parameter estimation will permit high-fidelity physical models for vehicle ground interaction as a basis for advanced control.

Situational Awareness

We are examining behavior-based situational awareness and probabilistic trajectory planning. The goal is to reclassify vehicle operations as experienced in real traffic situations and to embed them into a probabilistic framework based upon behavior. Here, multi-model adaptive estimation techniques reflecting a set of standard maneuvers are employed to arrive at a risk map. In return, this allows for a novel transition from path planning to behavior planning.