The Aerospace Robotics and Embedded Systems group performs research in the area of real-time planning and control for mobile cyber-physical systems (i.e., systems endowed with computation, communication, sensing, mobility, and actuation capabilities), with particular emphasis on autonomous vehicles, mobile robotics, and transportation systems.
Problems we deal with present numerous challenges, such as:
- Complex dynamics: We are interested in vehicles, such as aircraft and off-road vehicles, characterized by fast, uncertain, and nonlinear dynamics, subject to substantial disturbances, and to stringent safety requirements. Moreover, in order to perform complicated missions, and/or ensure safe interactions with other vehicles (including human-controlled ones), precise logic-based protocols must be followed, including, e.g., rules of the road, or rules of engagement. Our objective is to develop methods to design real-time control algorithms for these vehicles enabling high-performance behaviors, comparable to those achievable by expert human pilots or drivers. In addition, we aim at the design of high-confidence systems, which are guaranteed to operate in a safe, predictable, and well-understood manner, e.g., in an environment shared with humans and other vehicles.
- Sensing uncertainty: Robotic vehicles are able to perceive their immediate surroundings through sensors that often provide large amounts of noisy information, which must be processed and interpreted in real time to enable sound decision-making and motion control. Information about the environment beyond the immediate range of on-board sensors is generally unavailable, making it a necessity to be able to explore the environment and react to the sensed data. Conversely, mobility allows the rapid ad-hoc deployment of sensor networks able to monitor rapidly changing and geographically extended phenomena (e.g., weather, pollution). We investigate sensor data processing and motion control as fundamentally coupled problems, requiring capabilities beyond the current paradigms, often based on decoupling perception and control tasks.
- Large-scale distributed systems: The ability to exchange information with other vehicles can be exploited to increase the efficiency of interactions, possibly enabling cooperative behavior when compatible goals are sought. Our group develops algorithms governing interactions and cooperation in large-scale distributed systems, and methods to determine how performance, complexity, and robustness scale with the number of interacting agents.
In order to address the above challenges, the research in the group is highly interdisciplinary, requiring background and interest in several disciplines, including control theory, algorithmic robotics, estimation and signal processing, game theory, machine learning, theoretical computer science, networking and information theory, and operations research. We emphasize an analytical approach, aimed at understanding the fundamental issues in the problems of interest, and at the development of new, computationally-efficient planning and control methods, with provable correctness and performance guarantees. Our methods are typically demonstrated by simulation or on robotic test beds developed by the group and/or our collaborators.
The activity of the group is supported by a number of government sources – primarily NSF, AFOSR, and ONR – as well as industrial partners such as Aurora Flight Sciences.