Dynamic Models for Real-Time Autonomous Decision Making in the Real World

Successful applicants will join VinUniversity as Master’s students at the College of Engineering and Computer Science (CECS), VinUniversity

Project Information 

In recent years, the demand for deploying autonomous robotic systems in complex and high-risk environments, such as natural resource extraction, industrial maintenance, and demining operations, has been rapidly increasing. However, most existing systems still rely heavily on human teleoperation, limiting operational efficiency and increasing costs.

Real-time decision-making plays a critical role in enhancing robotic autonomy. Nevertheless, such systems face multiple sources of uncertainty, including noisy sensors and actuators, incomplete knowledge of the environment, and uncertainties in action outcomes and robot states. Interactions with other systems, including legacy teleoperated robots, further add to the complexity.

This project focuses on developing decision-making approaches under uncertainty, incorporating both stochastic uncertainty from the environment and epistemic uncertainty arising from the simulation models themselves. Instead of relying on a fixed model, the project proposes the use of dynamic models, where the level of detail and accuracy is adaptively adjusted based on the criticality of the state and situation.

This approach enables the use of simpler, faster models in less critical scenarios, while switching to more precise and fine-grained models when evaluating critical situations. As a result, the system can better estimate the expected outcomes of actions, leading to improved decision-making performance in real-world environments.

Principal Investigator: Asst. Prof. Leandro Soriano Marcolino

Research Objectives 

The project focuses on the following key objectives:

(i) Develop decision-making methods under uncertainty, integrating both environmental uncertainty and model-based uncertainty.

(ii) Design dynamic models capable of adapting their precision and the granularity of state and action spaces according to the criticality of situations.

(iii) Integrate dynamic models into sampling-based approaches to improve the estimation of expected action outcomes in critical scenarios.

(iv) Investigate interactions between autonomous robotic systems and legacy teleoperated systems in complex real-world environments.

Through these objectives, the project aims to enhance robotic autonomy, improve operational safety and efficiency, reduce costs, and minimize environmental impact in industrial applications.

Applications and Impact 

The outcomes of this project have strong potential for real-world applications, including:

• Nuclear decommissioning (in collaboration with UKAEA through partners in Manchester University)

• Maintenance of natural resource extraction facilities (in collaboration with Petrobrás and Vale, through our partners in Vale Institute of Technology and University of São Paulo)

The novel methods and algorithms developed are expected to advance the deployment of autonomous robotic systems in complex real-world environments.

Project Contact 

For further information, please contact Dr. Leandro Soriano Marcolino via email [email protected]

Graduate Admissions Contact 

• VinUni Graduate Admissions
•  0978 549 846
• [email protected]

Learn more about Master’s, PhD programs, and scholarships at: VinUni Graduate Research Excellence Program

#VinUni #VinUniversity#VinUniGraduateResearchExcellenceProgram #VinUniPhD #VinUniMasters#VinUniResearchProjects