Automation department seminar

Title : Safe stabilization using a Lyapunov barrier function for non-smooth control

Summary : In this talk, we address the problem of safe stabilization, whose control objective is to return the system state to the origin while guaranteeing the avoidance of dangerous regions in the state space.

Existing approaches, based on smooth control Lyapunov barrier functions, often fail to guarantee the existence of a feasible controller.

To overcome this limitation, we propose a framework of non-smooth control Lyapunov barrier functions (NCLBFs) that guarantees the existence of safe stabilizing controllers. We develop a systematic methodology for constructing NCLBFs and the corresponding feedback control laws for systems with several bounded dangerous regions.

Theoretical and numerical results demonstrate the effectiveness of the proposed approach and highlight its advantages over existing methods based on smooth functions.

Biography : Jianglin Lan is a Senior Lecturer at the James Watt School of Engineering, University of Glasgow, and an Honorary Research Fellow at Imperial College London. He received his PhD from the University of Hull in 2017. His research interests include safe autonomy, intelligent transport systems and robotics. He is editor-in-chief of the International Journal of Adaptive Control and Signal Processing, associate editor of IEEE Transactions on Intelligent Transportation Systems and a member of IFAC's SAFEPROCESS technical committee.

Title : Direct initialization by "sparse" method for visuo-inertial stereoscopic odometry

Summary : Existing methods for visuo-inertial stereoscopic initialization rely primarily on intermediate variables, such as feature matches and camera poses, rather than the original image data.

Calculating these variables using feature tracking and the Structure-from-Motion (SfM) method inherently introduces errors. To remedy this problem, we propose a direct initialization method for visuo-inertial stereoscopic odometry, which directly links the original image intensities and initial parameters, bypassing conventional intermediate variables.

The SfM method is based on the original image data.

More specifically, we introduce a prediction function to compute the corresponding points from the initial parameters and formulate an optimization objective that minimizes the sparse photometric error without requiring feature tracking or SfM.

We further develop an approximation method for two-image initialization, which demonstrates efficient performance even with minimal image data. Extensive experiments confirm that our method achieves superior performance in terms of both estimation accuracy and initialization success rate.

We also develop an approximation method for two-image initialization, which demonstrates efficient performance even with minimal image data.

Biography : Junyin Qiu is currently working towards a PhD at the James Watt School of Engineering, University of Glasgow, under the supervision of Dr Jianglin Lan. He obtained a master's degree in applied computer technology from Peking University in 2023, and a bachelor's degree in electronic information engineering from Sichuan University in 2020. His research interests include robotic localization and mapping, 3D computer vision and multisensory fusion.

Title : Direct initialization by "sparse" method for visuo-inertial stereoscopic odometry

Summary: This seminar explores different approaches to the design of observer-based state feedback controllers, especially suited to compensating for unknown external disturbances.

We focus on observer structures capable of simultaneously estimating unmeasured plant states and additive disturbances. These estimates are leveraged to develop robust, disturbance-tolerant control laws that maintain performance despite model uncertainties.

We focus on observer structures capable of simultaneously estimating unmeasured plant states and additive disturbances.

In order to demonstrate the practical utility of these methodologies, the presentation will outline some experimental applications and validations, highlighting the transition between theoretical design and concrete implementation.

Biography: Luciano Frezzatto obtained a bachelor's degree in computer engineering, a master's degree in mechanical engineering and a PhD in electrical engineering from the University of Campinas, Brazil, in 2009, 2011 and 2017 respectively. From 2015 to 2016, he interned at the Department of Mechanical and Aerospace Engineering at the University of California, San Diego, USA. He was an associate professor at the Federal University of Minas Gerais, Brazil, from 2018 to 2025. He is currently Associate Professor (Professor Doutor) at the Polytechnic School of the University of São Paulo (USP), Brazil. His research interests include robust control theory, LPV and fuzzy systems, fault-tolerant control and their applications.

Title : MPC with random constraints assisted by reinforcement learning for autonomous manure cleaning in dairy barns

Summary : Autonomous cleaning of dairy barns requires a robot to perform cleaning tasks efficiently while ensuring safe interaction with freely moving cows. This presents a challenge, as cows can stop, walk slowly, block narrow passages or move unpredictably, making fixed distance safety checks either too cautious or insufficiently safe.

This presentation proposes a chance-constrained predictive control framework, assisted by reinforcement learning, for autonomous cleaning of dairy barns.

The reinforcement learning policy provides task-level cleaning instructions, while the random-constrained MPC layer projects the robot's action into a probabilistically safe control space despite uncertain cow movements.

To further improve the trade-off between safety and efficiency, the method incorporates behavior-aware safety margins, which adapt according to the direction of cow movement and the risk of interaction.

Simulation results with 15 cows and 200 random trials show that the proposed controller improves mission efficiency and manure collection performance while maintaining safe interaction between cows and robot. Title: Predictive control with random constraints assisted by reinforcement learning for autonomous manure cleaning in dairy barns

Biography : Wei Wei is currently working towards a PhD at the James Watt School of Engineering, University of Glasgow, under the supervision of Dr Jianglin Lan. He received his master's degree in robotics and artificial intelligence from the same university in 2023. His research focuses on safe learning, control and planning for autonomous dairy cleaning robots. He is particularly interested in reinforcement learning, model-based predictive control, probabilistic safety and cow-robot interaction.