In recent years, a true revolution is taking place, in the way intelligent machines and robots operate in new, previously unseen, environments and interact with human operators. While in the past robots were primarily found in industrial settings, nowadays autonomous and semi-autonomous robots and systems can be found almost everywhere. This new generation of intelligent autonomous systems will interact even more closely with humans and will help them in their daily lives whether this is work, leisure, and by taking care of many mundane domestic tasks. But world is a messy place. There is a huge difference between a robot operating inside an enclosed “cage†on a factory floor that repeats the same task over and over again, and a robot that needs to navigate in an office environment, in a hospital, or on the highway, where uncertainty and unpredictability dominate. This project will develop new algorithms that run inside the “brain†of these autonomous systems to enable them achieve optimal decision-making, thus increasing their reliability, predictability, performance, and fail-safe operation in the presence of uncertainty and under limited information. Self-driving vehicles, anthropomorphic robots, aerial drones, manufacturing automation systems, and precision surgical instruments among others, will all benefit from the results of this research. Although motivated by robot navigation problems, this project addresses a more fundamental problem in artificial intelligence and thus has a much broader applicability. All applications where a “minimum-energy†path is to be found, e.g., crack propagation in structures, protein folding, data retrieval in high-dimensional spaces, etc., will benefit from the results of this project.
This project will leverage techniques from randomized graph representations and methodologies from stochastic optimal control theory, and will combine the two in novel ways, in order to mitigate uncertainty and unpredictability during planning and decision-making for high-dimensional autonomous robotic systems. The specific research activities to be undertaken in this project are: First, randomized graphs will be used to obtain efficient abstractions of the environment by avoiding non-scalable grid-based techniques, along with the application of new uncertainty propagation techniques developed by the investigator to solve efficiently planning problems in high-dimensional spaces. Second, optimal feedback strategies for stochastic systems will be developed by utilizing the recent theory of forward/backward stochastic differential equations, along with the incorporation of hierarchical and randomized approaches to better explore the search space. Finally, this project will take advantage of recent advances from Machine Learning (ML) and the use of prior experience gained during previous similar instances of the problem to expedite optimal search during runtime. The experimental validation of the theory will take place in the investigator’s lab and will involve both graduate and undergraduate students. The results of this research will be disseminated to the community by journal and conference publications and by securing summer internship opportunities for the students to transition the results of their work to real-life engineering problems.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
