ENGLISH/JAPANESE
wabian img Biped humanoid robot group

WABIAN-2R (2006-)
(WAseda BIpedal humANoid No.2 Refined)
 * Purpose
 * Previous research
    * Overview
    * Application as a human
       motion simulator
    * Application in real
       environment
 * Collaborative research
 * Future work
 * Acknowledgements
 * WABIAN-2 (2005)
 * WABIAN-2/LL
    (2003-2004)
 * Development of robots in
    Kato/Takanishi Laboratory
 * Papers

‘ RoboSoM -A Robotic Sense of Movement-
"RoboSoM -A Robotic Sense of Movement-", a joint research project with Italy, France and Portugal, was accepted to the European Seventh Framework Programme (FP7) with the highest score!
RoboSoM


 * Purpose

We have two purposes to develop a biped humanoid robot.
The first one is to develop a robot that would be a human's partner,
while the second one is to develop a human motion simulator.

1. Biped humanoid robot as a partner

Humanoid_Robot

Cooperation with humans in various fields
  The robot industry thus far focused on developments of industrial robots, for highly specified and constrained applications. The social demands and trends, however, push robotics into new areas, like medical treatment, life rescue, entertainment etc. It is just a matter of time, when human beings will find robotic assistance necessary in their daily life.
  In order to accomplish these task, robots will have to be able to move in indoor and outdoor conditions. Biped humanoid robots are best suited for this application.


2. Biped humanoid robot as a human motion simulator

walk-assist_machine walk-assist_machine

Application as a human motion simulator
  In the near future Japan is becoming an aging society. A number of the seniors generally suffer from disabilities on their lower limb, and a demand for the development of welfare equipments increasing. However, conventional evaluation methods for the equipments depend on human body measurements, and there are problems of the reproducibility, measuring accuracy, and human safety during the experiments,in particular.
  Therefore, our group has suggested applying a biped humanoid robot as a human motion simulator. That means a robot which can perform equal exercises with human beings tests welfare equipments under development in place of a human being, and provides quantitative data. We expect this new evaluation method would be safe, quantitative and versatile.

-Page_TOP-
 *  Previous research
 *  Overview

* Specification

Humanoid_Robot Humanoid_Robot
Height mm 1500
Weight kg 64 (with batteries)
Degrees Of Freedom (DOF)
Leg: 6~2
Foot: 1~2 (passive)
Waist: 2
Trunk: 2
Arm: 7~2
Hand: 3~2
Neck: 3
Total: 41
Sensors 6-axis force/torque sensor
Photo sensor
Magnetic encoder
Gyro sensor
Actuators DC servo motor
Reduction mechanism Harmonic drive gear
Timing-belt/pulley
Batteries Li-ion battery
Drawing DOF configuration

Front

Right
  WABIAN-2R has been designed with 1.5m in height, and 64.5kg in weight. In order to mimic human movements, the robot has 41 DOFs and the movable range of the joints designed in reference to human's one.
WABIAN-2R (2009)

  The computer mounted on the trunk controls the motion of WABIAN-2R. It consists of a PCI CPU board and PCI I/O boards. As the I/O boards, HRP interface boards (16ch D/As, 16ch counters, 16ch PIOs), and 6-axis force/torque sensor receiver board are mounted.
  The operating system is QNX Neutrino ver. 6.3. The drive system consists of a DC servo motor with an incremental encoder attached to the motor shaft, and a photo sensor to detect the basing angle. Also, each ankle has a 6-axis force/torque sensor, which is used for measuring Ground Reaction Force (GRF) and Zero Moment Point (ZMP).
Control system

-Page_TOP-
 * Application as a human motion simulator

* 2-DOF waist mechanism  (2003)

pelvis_DOF
2-DOF model for the waist mechanism
  Because of the mechanical constraint, most of the biped robots perform a bent-knee gait (walk with the knees bent during walking). On the other hand, researches on human gait's analysis show that the pelvis motion plays a significant role in human's gait. Taking into account the results of these researches, our team developed a 2-DOF (roll, Yaw) waist. By using the redundant mechanism, the robot is able to perform a stretch-knee gait as well as human being does.

›Knee-bent walking
Conventional walking
(mpg/13[s]/2.3[MB])

Walking with constant waist height
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]


›Knee-stretch walking with waist movement
Stretch walking 1
(mpg/13[s]/2.3[MB])

Walking with the knees stretched
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]
Stretch walking 2
(mpg/12[s]/2.0[MB])

Walking with the knees stretched on tiled floor
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]
Stretch Walking 3
(mpg/12[s]/2.1[MB])

Walking with the knees stretched outdoor
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]


* Walking experiments with a walk-assist machine  (2005)

  As one of the applications as the human motion simulator, we carried out walking experiments with a walk-assist machine. Generally, an armrest of a walk-assist machine is set to equal height of a user's elbow. And it is known that the severer his/her disability is, the lower we should set it.
  Measuring the current value of the motors at the knee joints and the forces & torques which are applied to the arms and legs, we are able to calculate the energy consumption at knee joints and the load to arms. Then we found the fact with quantitative data that the lower we set the height of the armrest, the less the load of knees is.
  This is a good example that human motion simulator can evaluate empirical facts quantitatively.   From the results of these experiments, we confirmed that we can evaluate welfare equipments quantitatively with a human motion simulator.
WABIAN-2 with a walk-assist machine

›Walking experiments with a walk-assist machine  (2005)
Simulations of walking assisted by a walk-assist machine
(mpg/8[s]/1.5[MB])

Knee-stretched walking with a walk-assist machine


* Foot with a passive toe joint  (2006)

  A human foot has a complex structure consisting of a lot of bones. Based on the result of gait analysis with motion capture system, we have developed a new foot which has a passive toe joint.
  Some of the clinical reports on toes' function indicate that toes do not produce thrust during steady gait. Since, in the current stage of our research, we focus on in steady walking, we have decided to develop a foot with a passive toe joint. Its main advantages are lightness and no necessity of a complex control.
Foot with a passive toe joint

›Human-like walking (Knee-stretched walking with heel-contact and toe-off motion)
Knee-stretched, heel-contact and toe-off motion 1
(mpg/13[s]/2.2[MB])

Indoor
Walking cycle: 1.0[s/step]
Step length: 350[mm/step]
Knee-stretched, heel-contact and toe-off motion 2
(mpg/11[s]/1.9[MB])

Indoor
Walking cycle: 1.0[s/step]
Step length: 500[mm/step]
Knee-stretched, heel-contact and toe-off motion 3
(mpg/13[s]/2.2[MB])

Outdoors
Walking cycle: 1.0[s/step]
Step length: 350[mm/step]
Knee-stretched, heel-contact and toe-off motion 4
(mpg/11[s]/1.9[MB])

Outdoors
Walking cycle: 1.0[s/step]
Step length: 400[mm/step]


›Detail of the heel-contact and toe-off motion
Detail 1
(mpg/8[s]/1.3[MB])

Comparison of the foot motions between the robot and human
Detail 2
(mpg/12[s]/2.1[MB])

Detail of the foot motion while walking


* Emulation of a disabled person's gait  (2007)
  As an example of the robot's applications, we had emulation experiments of a disabled person's gait. We used the walking data of a disabled person, recorded with motion capture system. Optimazing a part of walking parameters with Genetic Algorithm (GA), the robot achieve a good balance between walking stably and reproducting the gait. As the result, we confirmed the effectiveness from the stick diagram in the sagittal plane and the comparison of each joint angle.

Emulation of a disabled person's gait


›Emulation of a disabled person's gait  (2007)
The emulation of a disabled person's gait
(mpg/15[s]/2.7[MB])

Emulation of a disabled person's gait


* Foot mechanism capable of imitating human arch structure  (2009)



        Foot with human-like arch structure
   What is distinct in human beings compared to animals is that our feet are equipped with arch like structure. Researchers in human science field agree that the arch in human foot has significant influence to the gait, however it has never been fully investigated.
  The goal of this research is to implement the arch containing foot in the humanoid robot and further elaborate the influence of the foot arch to the gait of human being. The stiffness of the arch changes depending on the stage of the gate. It is thought, that the stiffness is very low in the flat foot phase, it rises with movement of the body forward and reaches its highest value at the toe-off phase.
   Experiments we performed after completing the development, proved that the arch structure absorbs the impact generated when sole fully contacts the ground.


* Swing pattern modification control  (2009)



Outline of the swing pattern modification control.
   Until recently, we used in our robot a posture error compensation control to perform the stable gait. Since the implementation of the human like foot, much narrower than the foot used thus far, the control algorithms turned out to be not able to generate the stable gait. To solve this problem we implemented new control strategy which, unlike the previous one, introduces a positive posture error. When the robot starts falling towards the inner side, the swing leg abducts before touching the ground and stabilizes the posture. The cycle is repeated for the opposite leg, simultaneously realizing the pattern.
  By implementing this control strategy, we succeeded in generating stable gate with heel contact and toe-off phases in humanoid robot equipped with feet containing human-like arch structure.

›Walking with heel contact and toe-off motion of WABIAN-2R equiped with feet containing human-like arch structure.
Overal view
(mpg/12[s]/1.9[MB])

Walking cycle: 1.0[s/step]
Step length: 450[mm/step]
Foot detailed view
(mpg/12[s]/1.9[MB])

Walking with changing arch

-Page TOP-
 *  Application in real environment

* Experiments in public roads in Fukuoka city  (2006-2007)

  For expansion of applied fields, we experimented in the environment where a human being lives. In particular, in public roads in Fukuoka city that was the one of the special zones for robot development and test, to confirm problems with the present walking ability, and to develop new control systems.
  This experiment was commissioned by the Robotics Industry Development Council (RIDC).
Experiments in public roads in Fukuoka city


* Power supply system  (2007)

  WABIAN-2R can use an external power supply and Li-ion batteries inbuilt in the body. On the other hand, we also use the external power supply when tuning-up the robot, and the inbuilt battery when performing experiments. WABIAN-2R can switch smoothly between the external power supply and the batteries, to realize effective experiments as a human motion simulator.
  Furthermore, WABIAN-2R can charge the batteries when operating with the external power supply.
Effective power supply management


* Foot mechanism adapting to the ground  (2008)

  In a real environment like outdoor road, bumps and slopes on the ground make walking unstable. WABIAN-2R can continue walking with stabilization control described above to some extent. However, it falls, when it fails adapting to the ground with big unevenness.
  We are trying to improve the walking ability not only with the control but also with a new foot mechanisms. One of the approaches is to detect ground before touching, while another is to adapt to the ground passively with a spherical joint mechanism.
iajFoot detecting ground ibjFoot adapting passively


* Walking stabilization control  (2008)


Outline of the walking stabilization control

  The gravity and inertia force must be balanced in a robot walking pattern (joint angle data on every moment). However, there are many errors that are not considered when the pattern is generated like unevenness, and inclination of the ground, an error between the real robot and the robot model (multi-mass system) and bend of the frame. So, a consideration of only gravity and inertia is not sufficient to achieve stable walking.
  To solve this problem, WABIAN-2R modifies the pattern based on the data of the GRF and attitude during walking.

  In order to do that, we use a vertual compliance control that makes the landing impact smaller. And also, Landing Pattern Modification ControliLPMCj changes the position and posture of the foot to adapt to the ground. Posture control corrects the error of robot's posture. WABIAN-2R can walk on a road with big slope or bump up to 5[deg] (roll and pitch) or 20[mm] (only pitch) with these control methods.
  In practice, we showed the high performance in experiments outdoor at Fukuoka city - stable walking on a slope in a park and on an uneven sidewalk.

›Walking experiment outdoor
Tiled uneven ground
(mpg/13[s]/2.2[MB])

Bump 5mm  Slope 2deg
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]
Big slope
(mpg/13[s]/2.2[MB])

Slope 5deg
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]


›20mm bump in lab
Without detecting ground
(mpg/13[s]/2.2[MB])

Whole foot on 20mm bump
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]
With detecting ground
(mpg/13[s]/2.2[MB])

Toe & heel on 20mm bump
Walking cycle: 1.0[s/step]
Step length: 200[mm/step]


* Foot adapting to the uneven surface  (2009)

  Since outdoor roads are not only rough and inclined, but also undulated it is not possible to perform the stable gait with use of flat feet. To cope with outdoor roads irregularities we improved the foot mechanism created in 2008 and come up with solution allowing us to negotiate also undulated surfaces.
  By implementing four moving pins on the corners of the foot, we change from the surface contact to the four points ground contact. Moreover, by sensing the displacement of each of the pins, we sens the shape of the ground and optimize the foot placement. With this solution, we were able to realize stable walk on the undulated surface with 20[mm] high unevenness.
Foot able to sens the surface shape Foot adaptation

›Walking on rough terrain in laboratory conditions (2009)    ›Walking on inclined surface (2009)
Verification of foot adaptation to uneven surface
(mpg/14[s]/2.0[MB])

Height of obstacles: 5-20[mm]
Verification of foot adaptation to the inclined surface
(mpg/14[s]/2.0[MB])

Robot starts the gait from the flat surface and proceeds to the slope inclined: 7.0[deg]

› Walking outdoor on rough surface  (2009)
Rough and inclined surface
(mpg/13[s]/1.8[MB])

Undulation height: 0-15[mm]
Inclination: 8.0[deg]


* Device for repetitive application of disturbing force  (2009)



Disturbance generator
   In real environment, robots are often subject to an unpredictable source of disturbance, like collision with obstacles or human. Our group works on control system for humanoid robots that would be able to respond to external disturbances. During algorithms verification experiments performed thus far, the disturbing force was applied by human. It is almost impossible for human to apply the same disturbing force in exactly the same phase of the gait, which is necessary to evaluate the efficiency of the algorithm. To solve this problem we developed machine able to repetitively generate the same disturbing force in any chosen time of the robot's gait.


* On-line generation of Furrier Transform based gait patterns   (2009)



Flow chart


Calculation of compensation ZMP trajectory based on necessary compensation moment

  Until now, gait patterns for WABIAN-2R were generated off-line and realized on the platform with on-line short-period pattern modifications. The disadvantage of this method is that in some circumstances, it is not possible to maintain a long time stability without modifying the predefined gait pattern.   In order to solve this problem we implemented an on-line generation of gait patterns. With this technology the robot can maintain long time stability. Moreover, it enables us to modify the gait trajectory on-line (e.g. with use of joystick).

› Walking with online pattern generation
Walking with waist height modification
(mpg/24[s]/3.9[MB])

Walking cycle: 1.0[s/step]
Step length: 200[mm/step]
Walking controlled with joystick
(mpg/84[s]/14[MB])

Walking cycle: 1.0[s/step]
Step length: 200[mm/step]

-Page TOP -
 * Collaborative research

  "SHINPO" was developed in a collaboration of tmsuk Co., Ltd. and Atsuo Takanishi Laboratory for a part of the robot exhibition at Niigata Science Museum, which Uchida Yoko Co., Ltd. had contracted for.
  The 2-DOF waist enables it a more human-like walking style.

Referential web page;
    Uchida Yoko Co., Ltd.
    http://www.uchida.co.jp/global/
    Niigata Science Museum
    http://www.lalanet.gr.jp/nsm/english/index.html
    tmsuk Co., Ltd.
    http://www.tmsuk.co.jp/tmsuklab/

SHINPO


  "KIYOMORI" was developed in a collaboration of tmsuk Co., Ltd. and the Atsuo Takanishi Laboratory.
  The same as "SHINPO", the 2-DOF waist enables it a more human-like walking style.

Referential web page;
    "KIYOMORI" special web page
    http://kiyomori.jp/main.html

KIYOMORI

-Page_TOP-
 * Future work
Future Works Future Works   We are trying to make the robot's ability higher to move close to humans and to have experiments with welfare equipments.
  In the future, we will establish the method to evaluate medical and welfare devices. Finally, we would also like to realize the human motion simulator, which can apply it to not only rehabilitation but also various fields.
Future Work
-Page_TOP-
 * Acknowledgements
  A part of this research was commissioned by a Grant-in-Aid for the WABOT-HOUSE Project by Gifu Prefecture, "Project for the Practical Application of Next-Generation Robots" by The New Energy and Industrial Technology Development Organization (NEDO), The Robotics Industry Development Council (RIDC) and supported by Grant-in-Aid for Scientific Research (No.18360126) from the Ministry of Education, Culture, Sports, Science and Technology. This project is executed under Humanoid Robotics Institute, Waseda University.
  Special thanks to SolidWorks Japan K.K for the 3D-CAD contribution, to QNX SOFTWERE SYSTEMS for the OS contribution, to DYDEN Corporation for the wire and harness contribution, to KITO Corporation for the robot supporting crane contribution and to every cooperative companies, local governments and public agencies.
HRI Humanoid Robotics Institute, Waseda University
WABOT HOUSE WABOT-HOUSE Laboratory, Waseda University
TOYOTA Motor Corporation
The New Energy and Industrial Technology Development Organization  (NEDO)
The Robotics Industry Development Council  (RIDC)
SolidWorks Japan K.K.
QNX SOFTWARE SYSTEMS
DYDEN Corporation
KITO Corporation
tmsuk Co., Ltd.
-Page_TOP-
Takanishi Laboratory TOP Last Update 2010.01.08

Copyright(C) 2009 Takanishi Laboratory
All Rights Reserved.