MAEDA Lab

INTELLIGENT & INDUSTRIAL ROBOTICS

Maeda Lab: 2021–2022

Div. of Systems Research, Faculty of Engineering / Specialization in Mechanical Engineering, Dept. of

Mechanical Engineering, Materials Science, and Ocean Engineering, Graduate School of Engineering Science /

Interfaculty Graduate School of Innovative and Practical Studies (Applied AI) /

Dept. of Mechanical Engineering, Materials Science, and Ocean Engineering, College of Engineering Science,

Yokohama National University

Mechanical Engineering and Materials Science Bldg. (N6-5),

79-5 Tokiwadai, Hodogaya-ku, Yokohama, 240-8501 JAPAN

Tel/Fax +81-45-339-3918 (Prof. Maeda)/+81-45-339-3894 (Lab)

E-mail maeda[at]ynu.ac.jp

https://iir.ynu.ac.jp/

People (2021–2022 Academic Year)

• Dr. Yusuke MAEDA (Professor, Div. of Systems Research, Fac. of Engineering)

• Doctoral Students (Specialization in Mechanical Engineering, Dept. of Mechanical

Engineering, Materials Science, and Ocean Engineering, Graduate School of Engineering

Science)

– Yao DENG

– Reiko TAKAHASHI (JSPS Research Fellow)

• Master’s Students (Specialization in Mechanical Engineering, Dept. of Mechanical

Engineering, Materials Science, and Ocean Engineering, Graduate School of Engineering

Science/Interfaculty Graduate School of Innovative and Practical Studies)

– Suneet SHUKLA

– Hiroyuki IHARA

– Yasuaki TANAKA

– Takuya NAKATSUKA

– Yukari HIRAKI

– Qian LI

– Akitoshi SAKATA

– Akihide SUGA

– Naruya SUZUKI

– Kenta TAKAHASHI

– Yuta NAKANISHI

– Yoshiki TAHARA

• Undergraduate Students (Mechanical Engineering Program, Dept. of Mechanical

Engineering and Materials Science/Dept. of Mechanical Engineering, Materials Science,

and Ocean Engineering, College of Engineering Science)

– Dan KOBAYASHI

– Haruki KAMIKUKITA

– Hirotaka KONDO

– Kenta SAKAKI

– Mizuki SHONO

2021–2022 Maeda Lab, Yokohama National University

SLAM-Integrated Kinematic Calibration (SKCLAM)

SLAM (Simultaneous Localization and Mapping) techniques can be applied to industrial manip-

ulators for 3D mapping around them and calibration of their kinematic parameters. We call this

“SKCLAM” (Simultaneous Kinematic Calibration, Localization and Mapping). Using an RGB-

D camera attached to the end-effector of a manipulator (Fig. 1), we demonstrated successful

SKCLAM in a virtual environment (Fig. 2) and a real environment (Fig. 3) [1][2]. We are also

studying SKCLAM with spherical cameras [3] and stereo cameras [4].

References

[1] J. Li, A. Ito, H. Yaguchi and Y. Maeda: Simultaneous kinematic calibration, localization, and mapping

(SKCLAM) for industrial robot manipulators, Advanced Robotics, Vol. 33, No. 23, pp. 1225–1234, 2019.

[2] A. Ito, J. Li and Y. Maeda: SLAM-Integrated Kinematic Calibration Using Checkerboard Patterns, Proc.

of 2020 IEEE/SICE Int. Symp. on System Integration (SII 2020), pp. 551–556, 2020.

[3] Y. Tanaka, J. Li, A. Ito and Y. Maeda: SLAM-Integrated Kinematic Calibration with Spherical Cameras

for Industrial Manipulators, Proc. of JSME Conf. on Robotics and Mechatronics 2020 (ROBOMECH

2020), 2P2-B05, 2020 (in Japanese).

[4] Y. Nagatomo, J. Li, Y. Tanaka and Y. Maeda: SLAM-integrated Kinematic Calibration with a Stereo

Camera for Industrial Robots, Proc. of JSME Conf. of Manufacturing Systems Division 2021, pp. 77–

78, 2021 (in Japanese).

Fig. 1 Manipulator Equipped

with an RGB-D Camera

Fig. 2 SKCLAM in Virtual Environment

Fig. 3 Example of an Obtained 3D Map

2021–2022 Maeda Lab, Yokohama National University

Robot Teaching

Teaching is indispensable for current industrial robots to execute tasks. Human operators have to

teach motions in detail to robots by, for example, conventional teaching/playback. However, robot

teaching is complicated and time-consuming for novice operators and the cost for training them

is often unaffordable in small-sized companies. Thus we are studying easy robot programming

methods toward the dissemination of robot utilization.

Robot programming with manual volume sweeping We developed a robot programming

method for part handling [1][2]. In this method, a human operator makes a robot manipulator

sweep a volume by its bodies (Fig. 4). The swept volume stands for (a part of) the manipulator’s

free space, because the manipulator has passed through the volume without collisions. Next,

the obtained swept volume is used by a motion planner to generate a well-optimized path of the

manipulator automatically. The swept volume can be displayed with Augmented Reality (AR)

so that human operators can easily understand it, which leads to efﬁcient robot programming

[3] (Fig. 5).

References

[1] Y. Maeda, T. Ushioda and S. Makita: Easy Robot Programming for Industrial Manipulators by Manual

Volume Sweeping, Proc. of 2008 IEEE Int. Conf. on Robotics and Automation (ICRA 2008), pp. 2234–

2239, 2008.

[2] S. Ishii and Y. Maeda: Programming of Robots Based on Online Computation of Their Swept Vol-

umes, Proc. of 23rd IEEE Int. Symp. on Robot and Human Interactive Communication (RO-MAN 2014),

pp. 385–390, 2014.

[3] Y. Sarai and Y. Maeda: Robot Programming for Manipulators through Volume Sweeping and Augmented

Reality, Proc. of 13th IEEE Conf. on Automation Science and Engineering (CASE 2017), pp. 302–307,

2017.

Manual Volume Sweeping

Swept Volume as a Part of Free Space Motion Planning within Swept Volume

(a) Programming Overview

(b) Manual Volume Sweeping

Fig. 4 Robot Programming by Manual Volume Sweeping

Fig. 5 AR Display of Swept Volume and Planned Path

2021–2022 Maeda Lab, Yokohama National University

View-Based Teaching/Playback

We developed a teaching/playback method based on camera images for industrial manipulators

[1][2]. In this method, robot motions and scene images in human demonstrations are recorded to

obtain an image-to-motion mapping, and the mapping is used for playback (Fig. 6). It can achieve

more robustness against changes of task conditions than conventional joint-variable-based teach-

ing/playback. Our method adopts end-to-end learning through view-based image processing and

therefore neither object models nor camera calibration are necessary. We are improving our view-

based teaching/playback by using range images (Fig. 7) and occlusion-aware techniques for more

robustness [3]. For application to force-control tasks, visualization of force information based on

photoelasticity (Fig. 8) is under investigation [4]. We are also trying to integrate reinforcement

learning with the view-based teaching/playback for reduction of human operations for teaching

[5].

References

[1] Y. Maeda and T. Nakamura: View-based teaching/playback for robotic manipulation, ROBOMECH J.,

Vol. 2, 2, 2015.

[2] Y. Maeda and Y. Moriyama: View-Based Teaching/Playback for Industrial Manipulators, Proc. of 2011

IEEE Int. Conf. on Robotics and Automation (ICRA 2011), pp. 4306–4311, 2011.

[3] Y. Maeda and Y. Saito: Lighting- and Occlusion-robust View-based Teaching/Playback for Model-free

Robot Programming, W. Chen et al. eds., Intelligent Autonomous Systems 14, pp. 939–952, Springer,

2017.

[4] Y. Nakagawa, Y. Maeda and S. Ishii: View-Based Teaching/Playback with Photoelasticity for Force-

Control Tasks, W. Chen et al. eds., Intelligent Autonomous Systems 14, pp. 825–837, Springer, 2017.

[5] Y. Maeda and R. Aburata: Teaching and Reinforcement Learning of Robotic View-Based Manipulation,

Proc. of 22nd IEEE Int. Symp. on Robot and Human Interactive Communication (RO-MAN 2013), pp.

87–92, 2013.

camera

robot

object

human instruction

(a) human teaching

mapping

image

robot

motion

(b) image-to-motion mapping

camera

robot

object

mapping

Fig. 6 Outline of View-Based Teaching/Playback

0 100 200 300 400 500

frame number

Grayscale

Images

Light

OFF

Range

Images

Fig. 7 Switching between Grayscale and Range

Images for View-Based Teaching/Playback

Fig. 8 View-based Teaching/

Playback with Photoelasticity