My research includes theoretical and rigorous empirical methods of study to provide a solid foundation for this emerging field. I emphasize both analytical modeling and experimental work in the lab so systems are built and tested and models are validated. The applications for intelligent embedded systems are wide-ranging; from intelligent homes, office and production environments, to security and reconnaissance.
Listed below are my specific research contributions and plans for future research. The detailed references for the papers cited below are available online as are the names of my principal collaborators on these works.
A central problem in intelligent embedded systems is localization (see J1). System autonomy often depends on the ability of each node to maintain a reasonable sense of its location relative to its environment. This ability is central to the unattended, long term functioning of the system.
I have recently begun investigating localization strategies for wireless nodes which are portable. These nodes (such as laptops, PDAs, sensors) move because a human (or other) agent transports them. Traditional robotics techniques which depend on inertial sensing on the moving object, do not work here. I have proposed a technique based on communication to localize such nodes. The basic idea is to treat the localization problem as an estimation problem where the measurements are noisy wireless signal strengths from several transmitters in the environment. Since the computational framework used is a Kalman filter, this formulation is a natural extension to the mobile robot node localization algorithm discussed below. I am currently working on a combined formulation which encompasses a variety of mobile nodes.
I am interested in the development of efficient, low-overhead estimation algorithms for mobile robot localization. In a collaborative project, I have proposed a Kalman filter-based algorithm to accurately localize mobile robots of varying morphology. This algorithm (and extensions) has been tested on several platforms including NASA/JPL planetary robot rovers and several indoor and outdoor wheeled mobile bases in the robotics laboratory at USC. The results appear in papers C5-C8. The algorithms have also been applied to estimate the state of aerial robots such as the autonomous robot helicopter at USC.
In concert with localization, mobile nodes need to formulate and maintain a representation of the world around them. In my work I have focussed on techniques which allow robots to autonomously build maps of their environment. With my students, I have developed algorithms for robot mapping which run in real time. Moreoever these algorithms are truly distributed, in that they can be used by multiple robots exploring an environment. This work relaxes the constraint which requires all robots to share a common coordinate system, or even to know each others' locations. The underlying map representation is a planar graph with directed, weighted edges. Map fusion across robots uses a fast topological graph matching algorithm. The results of this work appear in paper S3 and are currently being written up for submission to an archival journal. We are currently working on an extension of this algorithm which allows probabilistic matching between topologies.
Spatially distributed and networked embedded systems offer the promise of fault tolerance due to the absence of a centralized controller. However autonomously mobile systems, might be significant sources of faults within such networks if they are not reliably designed.
In collaboration with my students I have developed an architecture which utilizes traditional estimation algorithms and probabilistic techniques in a powerful hybrid architecture to detect, isolate, classify and rectify faults in mobile robot systems. This architecture has been implemented and improved in the course of several experiments with encouraging results (see C4, C10-C11). This is a fertile area of research in mobile robotics with several applications. In particular, under NASA/JPL support we are testing our techniques on small robot rovers for planetary exploration. The same architecture is being used (under DARPA support) to design and develop groups of robots with sufficient autonomy and fault tolerance to assist humans with reconnaissance and surveillance missions.
I am particularly interested in robust, distributed algorithms for mobile embedded systems which scale to large group sizes. A recent focus has been to develop biologically-inspired (ant-like) algorithms for mobile robots (C3). The strategies developed in this research are based on local computation and result in robust behavior. A recent result on conflict resolution (C1) allows robots to resolve interference issues (where there is contention for space) using communication. This is becoming especially relevant for mobile robots which must co-exist successfully with other robots and humans if they are to be useful.
A significant aspect of my research has been the development of user interfaces for intelligent embedded systems which have applications to virtual and real environments. As distributed systems of increasing complexity become the norm, innovative interfaces are needed for people to exploit them.
I have been working on the design of a haptic interface for users to explore virtual worlds. Such an interface is built using so-called haptic displays (gloves, force-feedback joysticks etc.) that provide the human user a sense of touch. My research interests in this area are efficient dynamic modeling of the virtual world and force control algorithms for providing realistic feedback (see C2). This research is supported by the Integrated Media Systems Center (IMSC) at USC. The problem of appropriate force control for haptic displays which rely on the the human hand is closely related to understanding the mechanics of human grasping. I have worked previously on building a human grasp-inspired architecture for robot hand control (see C16-19 and B1). I am using similar tools in the analysis of how to render forces effectively on a haptic glove-based display.
The study of distributed, intelligent, embedded systems needs tools to make experimental work possible. Since the systems I am interested in building are also the objects of analysis (to improve their performance for example), a substrate for experimentation and analysis is necessary. In particular, scaling experiments are difficult to perform using physical implementations.
Simulated robots in simulated worlds are useful for rapid prototyping of robot behavior. I am interested in developing a sustained program of research to design and develop efficient tools for simulation of mobile robot dynamics and control systems to substantially advance the state of the art in mobile robot design. In my Ph.D. dissertation work, I emphasized statistical interpretations of dynamic models to achieve a meaningful correspondence with observation (see papers J1, J2, C12-C15 and T3-T6). This is often adequate to prototype robot behavior and do design evaluation.