Autonomous mobile robots can contribute to several real-world applications such as emergency response.  In many cases, successful deployment in practical settings is subject to various forms of uncertainty that may not be necessarily modeled and/or accounted for prior to robot deployment.  Examples of such forms of uncertainty include robot-environment interactions that can be proximal (e.g., aerodynamic effects when flying close to surfaces for aerial robots) or physical (e.g., contacts and impacts for legged robots traversing uneven terrain).  In addition, for some applications, the cost of deploying large robots may be too high, and hence comparatively smaller robots may be preferred.  Such small robots can be especially error-prone in the presence of uncertainty since the small size restricts sensor and computational payloads, which in turn limits the robot's perceptual and control capabilities.

The main goal of the project was to introduce a platform-agnostic, data-driven modeling framework to quantify uncertainty and subsequently exploit it via control for reliable robot navigation under uncertainty.  The specific aims were to: 1) extract dynamics using limited data for modeling uncertain systems; 2) synthesize uncertainty-aware model-based controllers based on derived reduced-order models; and 3) test and validate theoretical analysis and derived models and control algorithms with different types of robots.  

Successfully meeting the stated goal and specific aims, this project developed a range of new algorithmic tools focusing on: 1) data-driven modeling and control, 2) reactive motion planning and control, and 3) interactive navigation.  The project had three major contributions:

• One important contribution was toward enabling the use of Koopman operator theory in (mobile) robotic applications.  The Koopman operator is one data-driven method that, in contrast to many other competing methods, can be used for real-time robot learning.  This research contributed toward both theory and practice, and developed methods were tested and validated across different types of robots.  
• Another important contribution was the development of new theory and algorithms to realize collision-inclusive motion planning.  Many robots nowadays can collide with obstacles safely, and this new capability was leveraged to create a new class of motion planning algorithms to deliberately switch between collision avoidance and colliding modes when it helps improve a navigation metric (for example time-to-goal).  The proposed approach was tested and validated with both kinematic (i.e. wheeled robot) and dynamic systems (i.e. aerial robot).
• Lastly yet importantly, the project also helped support allied efforts focusing on soft robotic systems, and was pivotal to help build a basis toward a new research direction on soft/compliant robot navigation (with specific focus here on compliant aerial robots and soft legged robots).  Deriving accurate models for such soft robotic systems (most of which being of small scale) is very challenging, and this project offered a way to cast this lack of modeling information as a source of uncertainty in motion control design, and then adapt the previous aforementioned algorithms in the context of soft/compliant robot navigation.  Part of the effort included designing and fabricating new compliant aerial and soft legged robots, and developing reactive and interactive motion planning and control algorithms for them.

Taken together, these findings helped push forward the state in (mobile) robot navigation under the presence of uncertainty along multiple distinctive directions, and contributed to the creation of new research directions.  While some of the basic principles developed in this research were initially grounded on small robots, this project's methods were able to generalize to larger robots with limited sensing and noisy actuation as well as to the distinctive class of soft/compliant robots.  This is evidence attesting to the project's overall success.

Besides the foundational robotics research findings summarized above, the project helped support (fully or partially) the research of three PhD students (all of whom graduated with their PhDs), and one MS student (also graduated).  The project also enabled three undergraduate students from underrepresented minority groups to become involved in related research, by capitalizing on the student diversity at UC Riverside, a Hispanic-serving Institution, thus adding to the School's effort to broaden participation of under-represented groups.

Last Modified: 12/06/2023
Modified by: Konstantinos Karydis