Research

Research

Control Applications

Researchers at CISCOR have applied state-of-the-art control algorithms and technology to real-world applications. Examples of the advanced control concepts are the use of stochastic arithmetic for efficient implementation of control laws, adaptive fuzzy control, neural networks, Kalman filtering, and fixed-architecture robust control. The applications come from many different domains, including electric ships, energy control, and industrial process control.

Dynamic Modeling for Vehicles

Research in dynamic modeling and power modeling of vehicles is motivated by motion planning of autonomous vehicles. Dynamic models are needed for planning tasks for planning tasks such as involving undulating terrains with steep slopes, slippery terrains, and mixed-terrains. The appropriate use of dynamic models enables the development of time-efficient or energy efficient trajectories.

Flow Control

Flow control research is accomplished in conjunction with the Florida Center for Advanced Aero-Propulsion (FCAAP). Past research has involved using modern control to develop and implement robust, feedback controllers for counterflow thrust vector control. Current research focuses on the use of adaptive control to facilitate flow separation control in aircraft systems using microjet technology that has been pioneered by FCAAP researchers. This technology is expected to lead to increased control and reduced noise of hovering aircrafts and increased maneuverability of flying aircraft.

Legged Locomotion: Design and Modeling

Although modern wheeled and tracked vehicles can best the speed of any animal, they still cannot challenge animals versatility and agility over natural, rough terrain. Legged machines have long held the promise of achieving such agility, but only recently have robots such as RHex and iSprawl been built which are capable of running at multiple body-lengths per second and over unstructured terrain. This research has a particular focus on designing, dynamically analyzing, and manufacturing advanced concepts legged robots. A major goal is to develop multi-modal vehicles, i.e., vehicles that can crawl, walk, run, climb, and even swim. More information on this research can be found on the STRIDE Laboratory web site.

Motion Planning for Mobile Robots and Manipulators

The focus of our motion planning research is to develop algorithms that will enable both mobile robots and manipulators to operate intelligently in extreme environments or for extreme tasks. For mobile robots such as autonomous ground vehicles (AGVs), autonomous air vehicles (AAVs), or autonomous underwater vehicles (AUVs) an example of an extreme environment is a cluttered environment. This problem has been addressed using both reactive and deliberative planning algorithms. For AGVs extreme environments also include difficult terrains such as sand, ice and mud, and highly undulating terrains. For manipulators an extreme task is lifting objects that are so heavy that they cannot be lifted quasi-statically. A unifying feature of each of these latter problems is that they benefit from using a dynamic model in the planning process. Hence, a major focus of this research is the development and refinement of Sampling Based Model Predictive Control (SBMPC), a novel nonlinear MPC (NMPC) approach that enables motion planning with dynamic models as well as the solution of more traditional MPC problems.

Multi-Agent Cooperation

A major paradigm for cooperative task allocation is auction methods. An ongoing problem is to develop computationally efficient auction methods that can enable the task allocation to continuously improve when the communication and bandwidth are available for continuous auctioning. This research introduces the concept of Stochastic Clustering Auctions (SCAs) that enable this continuous improvement and have parameters that can be used to trade off computational and communication requirements with optimality.

Object Classification

Principal Component Analysis (PCA) is a major paradigm for the development of algorithms for object classification. This research has focused on the development of more computationally efficient PCA-based object classification algorithms and algorithms that enable the classification of occluded objects.

Teleoperation

Teleoperation can lead to instability or poor performance in the presence of communication delays. This research uses wave variables to enable the development of provably stable teleoperation control in the presence of these delays.

Terrain Classification

The primary focus of this research is on the classification of terrain surfaces to enable automated update of a terrain-dependent control system. This research is motivated by the need for efficient and safe control of autonomous ground vehicles (AGVs) on outdoor terrains such as grass, gravel, sand, asphalt and ice. Terrain classification algorithms have been developed using both vision sensors and proprioceptive sensing such as direct vibration measurements and slip estimates. The research has also developed a methodology for filtering misclassifications so that the control system does not change due to these misclassifications. Ongoing research also includes the application of terrain classification to electric powered wheelchairs (EPWs) as part of the development of a terrain-dependent EPW control system.

Terrain Dependent Control

Terrain dependent control systems are imperative for efficient and safe operation of autonomous ground vehicles (AGVs) that will operate in highly unstructured outdoor environments of varying and sometimes difficult terrains. This research approaches the problem of developing these control systems by quantifying and automating terrain-dependent driving rules that have been developed by expert off-road drivers. These rules include limits on the vehicle acceleration and turn radius when on a terrain such as sand and the benefits of following ruts on soft surfaces such as snow, mud or sand.