All Publications

This paper presents a robust velocity planning method for robotics and autonomous vehicles using deep reinforcement learning, offering scene-independent, efficient, and comfortable performance.

This paper presents a multi-modal model-based reinforcement learning approach for excavating challenging fragmented rocks. It leverages a multi-modal recurrent neural network (RNN) and a model predictive controller (MPC) to outperform conventional strategies, demonstrating superior predictive accuracy and effectiveness in a challenging task.

This paper presents a two-stage approach that integrates data-driven imitation learning and model-based trajectory optimization to enhance automated excavation techniques for excavators, resulting in a significant increase in excavation efficiency.

This paper outlines an approach for efficient trajectory tracking control of autonomous wheel loaders, employing nonlinear model predictive control for trajectory planning and a Linear Parameter Varying (LPV) model for enhanced computational efficiency, leading to better performance and reduced computational burden compared to conventional methods.

This paper presents a hierarchical planning system for autonomous excavators, enhancing task and motion planning, and demonstrating its effectiveness in real-world and simulated excavation tasks.

This paper introduces an Autonomous Excavator System (AES) designed for material loading tasks in challenging environments. Combining advanced perception and planning algorithms, AES demonstrates autonomous operation with efficiency comparable to human operators and robust performance in complex scenarios, achieving 24-hour continuous operation without human intervention.

This paper introduces a robot guide dog system designed to assist vision-impaired individuals in navigating environments. Using a novel human-robot kinematic model and a Model Predictive Control (MPC) algorithm, this wheeled ground robot can successfully guide users through narrow corridors, avoiding obstacles and enhancing their mobility.

This paper emphasizes the potential of autonomous excavators to enhance safety and productivity by replacing human operators in hazardous conditions within a multi-billion-dollar global market.

This paper presents a K Nearest Neighborhood learning-based method for wind speed estimation in vertical take-off and landing UAVs. It offers robust and accurate wind estimation without the need for detailed aerodynamic information, addressing challenges posed by rotor down-wash effects.

This paper presents a monocular vision-based solution for autonomous quadrotor UAV landing on a vessel deck under challenging sea conditions, using on-board sensors only. Experimental results with a Parrot AR.Drone platform confirm the solution’s accuracy and robustness.