An Auto-Adaptable Algorithm to Generate Human-Like Locomotion for Different Humanoid Robots Based on Motion Capture Data

An Auto-Adaptable Algorithm to Generate Human-Like Locomotion for Different Humanoid Robots Based on Motion Capture Data

The 2010 IEEE/RSJ International Conference on Intelligent Robots and Systems October 18-22, 2010, Taipei, Taiwan An auto-adaptable algorithm to generate human-like locomotion for different humanoid robots based on Motion Capture Data Luc BOUTIN, Antoine EON, Said ZEGHLOUL, Patrick LACOUTURE Abstract— The work presented in this paper deals with the the proposed method is demonstrated with the results on a generation of trajectories for humanoid robots imitating human small sized robot (HOAP-3) and on a human-like sized robot gaits captured with a motion capture system. Once the human (HRP-2) which is quite unusual in the literature. Compared motion is recorded, this one is modified to be adapted to the robot morphology. The proposed method could be used for to the computer graphic applications, the difficulties for the human-like robots of different sizes and masses. The generated humanoid robots is to respect the feet/floor contacts and the gaits are closed to the human’s ones while respecting the dynamic balance. To imitate the human locomotion captured, robot balance and the floor contacts. First the human joint this paper presents a new approach based on the control angles are computed from the markers coordinates and applied of feet motion and CoM (Center of Mass) trajectories. The directly to the robot kinematics model. Then, from this non- corrected motion, the trajectories of both feet and of the Zero maximum speed of the computed gait is found in order to Moment Point (ZMP) are generated respecting the constraints respect the ZMP (Zero Moment Point) criterion [7]. of floor contact and balance control. From this data, an inverse The motion capture process and the analysis of the human kinematic algorithm is used to compute the joint angles of the locomotion is presented in the next part. Then the principle robot according to the feet and ZMP trajectories. The results of the motion adaptation is detailed with its key points: the with the robot HRP-2 (AIST, Kawada Industries, Inc) and the small-sized humanoid HOAP-3 (Fujitsu Automation Ltd) are feet motion definition, the balance constraint, and the specific compared with the human motion. solver. The last part of this paper presents the results of the motion adaptation to the human-sized robot HRP-2 and to I. INTRODUCTION the small-sized robot HOAP-3. The idea to generate humanoid motions from human motion capture is mainly within three scientific fields: biome- II. HUMAN LOCOMOTION chanics, computer graphics and robotics. On this topic, A. Motion capture biomechanics deal with the analysis of pathological subject Thirty seven reflective markers (Fig. 1) are placed on the gait in order to characterize the walking pathologies and subject skin (Fig. 1). To collect kinematic data i.e. to record finally finding an individual functional rehabilitation method the trajectories of the markers, a motion capture system is or a surgical solution. Our motion capture method is based on used with six cameras located around the studied area. the experience of biomechanics community [1] especially for the choice of marker sets and segment (a limb considered as Fig. 1. The marker set — Human segments a rigid body) axis definitions. Computer graphics [2] focus on the use of motion capture and synthesis movement to generate three-dimensional realistic movements for virtual models including non anthropomorphic models as well. This is related to the present work since the problem is how to transform the motion of a human with a special size to a humanoid robot with different dimensions, masses and inertias. The gap between these diverse fields is beginning to be closed through the work of multi-disciplinary teams. For instance a dance recorded by motion capture was learned and reproduced by the Humanoid HRP-2 [3]. More recently Ankle, knee, elbow and wrist joint centers are considered HRP-4C imitates a human female walking and turning [4]. as located in the middle of their internal and external The Honda research institute proposed an online transfer of markers. This method is currently used by the biomechanics human motion [5] but only for the upper part of the robot community [8], [9]. The shoulder and hip joint centers are ASIMO. not easy to locate precisely. Two main methods are available The main contribution of this work is to adapt the computer for their location. The first one called predictive method graphic techniques for retargetting motion to new characters uses several characteristic limb dimensions of the subject [2], [6] to generate trajectories for humanoid robots of to locate the joints according to the markers [10]. However different sizes imitating a human motion. The relevance of joint angles and inverse dynamic results are sensitive to the location of joint centers, this method is not as accurate as Institut Pprime, CNRS, Universite´ de Poitiers, ENSMA, SP2MI, Bd Marie et Pierre Curie, BP30179, 86962 Futuroscope, FRANCE. the second one called functional method. This one is based [email protected] on a computation of joint center locations from a kinematic 978-1-4244-6676-4/10/$25.00 ©2010 IEEE 1256 analysis of imposed movements corresponding of ten cycles D. CoP and CoM trajectories of flexion, abduction and circumduction [11]. In the present To verify the gait event detection, two force platforms were approach the functional method is used in order to determine used. The collected data contain also the trajectory of the the joint centers and thus the joint trajectories. The number center of pressure (CoP) during gait. The force platforms of segments chosen is fifteen (Fig. 1) which is sufficient to provide the results from the beginning of the first single transpose the motion to a humanoid robot (feet, shins, thighs, support phase to the end of the second one (Fig. 3). During pelvis, thorax, arm, forearm, hands, head). Most of humanoid the double support phase, the CoP is computed from both robots are designed with this kinematic architecture [12]. platforms. The wrist and neck joint are not considered in this study. The hands and the head follow respectively forearms and 400 Speed: 0.84m/s (3.01km/h) the thorax. The segment axis are chosen according to the Single support International Society of Biomechanics’ recommendations 300 CoP CoM Right foot (ISB)[1] except for the feet in order to simplify the def- 200 y (mm) Left foot inition of angles. In the reference anatomic position, human CoP CoM 100 CoM CoP Double support model segment y-axis are pointing upwards, while x-axis are Single support 0 constructed to be close to the walking direction. 0 100 200 300 400 500 600 700 800 900 x (mm) − walking direction B. Human joint angles computation Fig. 3. Measured forces of the platforms (500Hz) — Subject center of To extract any segment movement in the reference coor- pressure and center of mass projection on the floor dinate system (called R0), at least three markers are needed to build the segment coordinate system. The rotation matrix During the single support phase, the CoP moves from Rot(Si/R0) of each segment Si is computed with respect to the back to the front along the main axis of the foot. The R0. Then the rotation of the segment i with respect to the co- CoP is slightly on the external part of the foot. After the − ordinate system of the segment i 1, the matrix Rot(Si/Si−1) single support, the CoP changes immediately its direction, is computed. Finally the yaw-pitch-roll sequence is used to because only the toes of the first foot are in contact with the identify the intersegmental angles called abduction, flexion ground. The CoP moves from external to internal toes. At and rotation. the same time, the heel of the second foot enters in contact C. Foot contact events with the force platform. During the double support phase, the The next step of the process is to compute the human CoP moves in “straight line” from the end of single support foot contact events. Indeed, to generate a feasible gait for position to the beginning of second single support. the robot, the single (one foot on the floor) and double The CoM trajectory is computed from anthropometric data support (both feet on the floor) phases must be defined. The [15]. The CoM projection on the ground follows a sinusoidal foot contact events: Initial Contacts (IC) and Toes take Off curve as it is described in many papers [16]. The amplitude of (TO), are identified only with the kinematic data. A recent these oscillations are small compared to the CoP transverse gait event detection algorithm called the high pass algorithm displacement. This gait is clearly dynamic and non quasi- (HPA)[13] is extended for the present application. Kine- static. At the beginning and the end of the single support matics of markers on the lateral malleolus and metatarsal phase, the CoM lies outside the sustentation polygon which are used (Fig. 2). It was necessary to modify this algorithm means that the static balance is not respected. because it was developed for human gait in a straight line III. MOTION ADAPTATION while more complex locomotion are transposed to the robot. A. The idea The modification is to use not only the marker displacements along one axis (walking lane axis in the original case) but the To imitate the captured human motion with a humanoid curvilinear abscissa in the horizontal plane. For a non cyclic robot, a first approach could be to control the robot as a gait with a quarter turn at the end, the two first IC and TO are marionette [17], imposing the human joint angles to the compared to the measurement of two force plates located on humanoid.

View Full Text

Details

  • File Type
    pdf
  • Upload Time
    -
  • Content Languages
    English
  • Upload User
    Anonymous/Not logged-in
  • File Pages
    6 Page
  • File Size
    -

Download

Channel Download Status
Express Download Enable

Copyright

We respect the copyrights and intellectual property rights of all users. All uploaded documents are either original works of the uploader or authorized works of the rightful owners.

  • Not to be reproduced or distributed without explicit permission.
  • Not used for commercial purposes outside of approved use cases.
  • Not used to infringe on the rights of the original creators.
  • If you believe any content infringes your copyright, please contact us immediately.

Support

For help with questions, suggestions, or problems, please contact us