Research & Invited Talks
Research in autonomous mapping, dynamic-environment perception, mobile manipulation, and intelligent robotic systems.
Neuromorphic Computing and Memristor-Based Devices in Intelligent Robotics
Guest Lecturer · University of Seoul
Seoul, South Korea
Invited by the engineering faculty to deliver a lecture to an international engineering cohort on neuromorphic computer architectures, memristor-based devices, and the role of specialized hardware in energy-efficient AI systems.
A Supervised ROS 2 Architecture for Autonomous Mapping: Modular Exploration-Management Framework
First Author · CNRS-AIST Joint Robotics Laboratory
Tsukuba, Japan
Presents a modular ROS 2 software architecture for autonomous mapping in unknown and dynamic environments. The exploration-management framework handles frontier evaluation, navigation-goal supervision, failure recovery, progress monitoring, and mission completion while remaining configurable for Nav2-compatible robots.
This paper presents a modular ROS 2 architecture for autonomous mapping in mobile robots operating within unknown and dynamic indoor environments. The proposed system integrates multimodal scan fusion, a static–dynamic obstacle-filtering module, SLAM-based map construction, frontier-based exploration, and Nav2-based navigation, coordinated by a program manager that governs the mission from distributed launch through coordinated termination. Beyond this supervised software system, the architecture is organized around a modular ROS 2 exploration-management framework that extends beyond frontier detection to govern candidate evaluation, navigation-goal supervision, failure recovery, and mission completion. Requiring only an occupancy-grid map, robot-pose information through TF, and a Nav2-compatible navigation interface, the framework may be adopted by compatible ROS 2 mobile-robot stacks through external configuration, without modification of its underlying implementation. The complete system is validated in simulation and on the physical CALL-M mobile manipulator.
A ROS 2 Static–Dynamic Obstacle Filtering Framework for Robust Mapping in Dynamic Environments
First Author · CNRS-AIST Joint Robotics Laboratory
Tsukuba, Japan
Develops a reusable preprocessing framework that separates persistent environmental structure from moving observations before mapping, while preserving dynamic obstacles for real-time navigation, tracking, and collision avoidance.
This work presents a modular ROS 2 framework for separating static environmental structure from dynamic obstacles in LaserScan-based robotic mapping and navigation. The framework processes unified multimodal range observations to distinguish persistent structures from transient measurements, providing a static scan for SLAM-based map construction and a dynamic scan for real-time obstacle avoidance. In addition, dynamic observations may be tracked to estimate obstacle motion while preserving compatibility with conventional ROS 2 LaserScan interfaces. The approach is designed as a reusable preprocessing layer for mobile robots operating in dynamic indoor environments, improving map consistency while retaining moving obstacles for navigation and collision avoidance.
An Integrated Perception, Navigation, and Manipulation Pipeline for Semi-Autonomous Door Opening with a Mobile Manipulator
Second Author · CNRS-AIST Joint Robotics Laboratory
Tsukuba, Japan
Presents an end-to-end mobile-manipulation pipeline combining door perception, LiDAR-based Nav2 approach, RGB-D handle localization, base alignment, pre-grasp positioning, grasp execution, and physical door opening on CALL-M. Evaluated in 20 real-robot trials; in the 10 nominal trials, upstream perception, navigation, localization, alignment, and pre-grasp stages succeeded 10/10, while complete door opening succeeded 8/10.
Autonomous door opening is a key capability for service robots. It requires integration of perception, navigation, localization, and manipulation. This paper presents a semi-autonomous door-opening pipeline for a mobile manipulator. The system requires manual initialization of the robot pose in RViz; it autonomously detects and approaches the door, estimates the 3D handle position, aligns the mobile base, performs the pre-grasp positioning, and executes handle manipulation and door opening. The system was evaluated on a real mobile manipulator in 20 trials: 10 nominal operation trials, 5 trials with varying initial robot-to-door distance, and 5 trials varying initial end-effector height. In nominal condition, door perception, navigation, 3D localization, alignment, and pre-grasp positioning succeeded 10/10, while complete door opening succeeded 8/10. The system successfully handled the initial robot-to-door distance in four of five trials. However, changes in the initial end-effector height reduced performance, completing only one of five trials. Failures occurred mainly in the contact and interaction between the handle and the gripper. Physical handle manipulation is the main limitation of the current system.