Research

Autonomous systems and imitation learning

My work in autonomous systems investigates how learning agents can acquire reliable behavior from experience and demonstrations. Recent directions include behavior cloning, generative adversarial imitation learning (GAIL), energy-based models, and diffusion-based policies, particularly for autonomous driving in urban environments.

Projects have explored perception and control using Bird’s-Eye View (BEV) representations, the CARLA simulator, and miniature autonomous vehicles. Current group activities and demonstrations are documented at GSAI.

Physics-informed modeling and control

An important research direction is integrating prior physical knowledge into machine learning, especially for dynamic systems governed by differential equations. This includes Physics-Informed Neural Networks for Control (PINC) and data-driven neural models for efficient simulation, system identification, and control.

Selected publication: Physics-Informed Neural Nets for Control of Dynamical Systems — Neurocomputing (2024).

Reservoir computing and recurrent neural networks

My earlier research has examined Echo State Networks, reservoir computing, online learning, and recurrent neural network models. Applications include autonomous robot navigation and localization, efficient modeling of nonlinear systems, and neural control of industrial processes.

Selected publication: On Learning Navigation Behaviors for Small Mobile Robots with Reservoir Computing Architectures — IEEE TNNLS.

Industrial applications

Applications of data-driven dynamical modeling include process monitoring, soft sensors, and modeling and control in oil and gas production. These activities complement my methodological research on recurrent and physics-informed machine learning.

For recent group projects, members, papers, and demonstrations, visit GSAI — Intelligent Autonomous Systems Group. For a complete record of academic publications, see Google Scholar or Lattes.