Robot Navigation in Crowded Environments


Project Description

One of big challenges in robotics research is in enabling robot navigation in crowded environments. As one key technical difficulty, in such environment, a robot would trap into a deadlock state where the robot is unable to advance to its destination, especially when it is surrounded by pedestrians and perceives them as obstacles which the robot is programmed to avoid collision against. This problem is widely known as a "freezing robot problem" and acts as a major hurdle in deploying robotic systems in real-world applications. We study this problem in the context of human-robot interaction and develop new algorithms that ensure deadlock-free robot navigation in pedestrian-populated environments. The project aims at developing both principles and algorithms to address the freezing robot problem: we begin by investigating how the human-robot interaction can be understood as a feedback interconnection of human and robot decision-making models and then identify under what specifications on the robot's decision-making model, the robot is guaranteed to be deadlock-free. We will explore tools from feedback control theory and game theory to find such specifications and develop robot navigation algorithms satisfying them. Ultimately, we will carrying out experiments using multiple mobile robot platforms to validate our framework in pedestrian-populated areas.
Program - Electrical Engineering
Division - Computer, Electrical and Mathematical Sciences and Engineering
Field of Study - Electrical and Computer Engineering

About the

Shinkyu Park

Assistant Professor, Electrical and Computer Engineering

Shinkyu Park

Professor Park's research interests are in the general areas of robotics, multi-agent decision making, and feedback control. His most recent research has been in design and control of multi-robot systems and related topics of game theory and feedback control systems, with applications to multi-robot learning and coordination.

Desired Project Deliverables

The goal of this project is to develop, implement, and test a deadlock-free robot navigation framework based on well-established multi-agent decision-making models. The students will get to learn many of tools designed for robot perception and navigation tasks and gain experience in designing and implementing their own algorithms into robot platforms. Then the students are expected to design new methods to predict pedestrian motions using data from on-board sensors and to maneuver the robot in crowded areas. The students are encouraged to work on research-oriented problems and expected to come up with their own creative solutions. They should have backgrounds in robot perception and motion planning, and they are expected have experience with robotics software and proficiency in programming languages (C/C++ and Python).