Objective:

The aim of the workshop is to share the state of the art of research on the motion planning for humanoid robots interacting 
with the real world. The project in the JST-CNRS program concerns "Robot motion planning and execution through online 
information structuring in real-world environment". Each presentation will include this aspect.

 

Organizers:

Eiichi Yoshida (AIST), Kensuke Harada (AIST), Kazuhito Yokoi (AIST),

Florent Lamiraux (CNRS) and Jean-Paul Laumond (CNRS)

 

Program:

 

9:00-9:05 Opening

 

9:05-9:40 Title: Modeling the Real World for Humanoid Motion Planning

Keynote speech: Prof. Katsu Yamane, Carnegie Mellon University, Disney Research Pittsburgh, USA

 

Abstract: Motion planning in the real world is all about finding the input sequence that optimizes the robot motion with respect to a given cost function 
before actually executing the motion.  There are four main issues here in the order of occurrence: 1) choosing the cost function to optimize, 
2) sensing the robot state and environment to initialize the search, 3) sampling in the search space, and 4) predicting future state resulting from sample 
inputs.  In this talk, I mostly focus on 1) and 4) for planning whole-body humanoid motions, along with a couple of application examples.  These issues 
are essentially the problem of how to model the real world including robot, human and environment.
 
Choosing appropriate cost function is not as trivial as it might appear, especially in applications such as entertainment robots where human perception 
of the robot motion is the primary concern.  In such cases, obviously some measure of "naturalness" should be included in the cost function, but formal 
definition of naturalness is still an open question. I will introduce our work on trying to understand the quantitative measure of naturalness through human 
motion analysis based on detailed musculoskeletal models.
 
Precise prediction of future states is also important because the cost function is evaluated based on those predictions.  It is also desirable to carry out 
the prediction efficiently.  I introduce two algorithms for this purpose: inverse kinematics to predict whole-body posture from given desired endpoint 
locations, and forward dynamics to predict the motion from given joint torques and environmental constraints.  I also present an example of applying 
the inverse kinematics algorithm to planning manipulation tasks where we use a motion capture library to improve the quality of prediction.

 

 

9:40-10:05 Motion and Task Planning for HRP2JSK Humanoid Robots

Kei Okada, Satoru Tokutsu, Toshiaki Maki, Yuto Mori, Ryohei Ueda, Shunnichi Nozawa, Kimitoshi Yamazaki, Masayuki Inaba, University of Tokyo, Japan

 

Abstract : Motion planning and task planning techniques are powerful tool for humanoid robots performing daily assistive tasks autonomously. They can generate complex behavior from simple and high level description of constraints or goals. One of the essential problems within this research field is the following: How to adapt planned behaviors into the dynamic real world environments with uncertainty.

In this presentation, we present motion planning and task planning integration for autonomous humanoid behaviors with failure recognition and recovery feature. Then we will introduce sensor-based and sensor-aware motion planning approach towards motion planning for situated robots.

 

10:05-10:30 Humanoid Motion Planning for Reaching, Grasping and Re-Grasping

Tamim Asfour*, Nikolaus Vahrenkamp*, Dmitry Berenson+, Markus Przybylski*, James Kuffner+ and Rüdiger Dillmann*

*Institute for Anthropomatics +The Robotics Institute University of Karlsruhe Carnegie Mellon University (CMU) Karlsruhe, Germany Pittsburgh, USA

 

Abstract: In this talk, we will present our work on motion planning for grasping and manipulation tasks on humanoid robots consisting of offline grasp planning for a five]fingered hand and several motion planning methods for single and dual arm reaching, grasping and re]grasping tasks.

Offline generated grasp candidates for the five]fingered hand and a given known objects are used as input for the planning algorithms. We present two probabilistically complete RRT]based algorithms for reaching, grasping and re]grasping 1) Jacobian]Pseudoinverse]based planning (J+ ]RRTs), which is an extension of the single]tree RRT]JT approach and combines the configuration space search of RRTs with multiple workspace goals given by the grasp candidates without the need for any inverse kinematics and 2) IK]RRT, which is a bidirectional RRT that samples IK solutions during planning. In addition we present extensions of both algorithms to deal with dual arm grasping tasks. Evaluation and experimental results both in simulation and on the humanoid robot ARMAR]III in a kitchen scenario will be presented.

 

10:30 -10:45 Coffee Break

 

10:45-11:10 Title: Whole Body Motion Planning for Humanoid Robots

Kensuke Harada and Fumio Kanehiro, Humanoid Research Group, National Institute of Advanced Industrial Science and Technology (AIST), Japan.

 

Abstract: In this talk, we will present several topics on motion planning for humanoid robots. First, we explain the simultaneous planning of the whole body motion and the biped gait. Then we discuss the quadratic programming method applied for the whole body motion planning. We further discuss the grasp planning of a multi-fingered hand attached at the tip of the humanoid robotfs arm. The effectiveness of the proposed approach will be confirmed by several simulation and experimental results.

 

11:10-11:35 Whole-Body Motion Planning of Human-Like Robots and Applications toVirtual Prototyping

Liangjun Zhang, Jia Pan, Dinesh Manocha, Dept. of Computer Science, University of North Carolina at Chapel Hill

 

Abstract: We give an overview of a whole-body motion planning algorithm for human-like robots. The planning problem is decomposed into a sequence of low-dimensional sub-problems. Our formulation is based on the fact that a human-like model is a tightly coupled system and uses a constrained coordination scheme to solve the sub-problems in an incremental manner. We also present a local path refinement algorithm to compute collision-free paths in tight spaces and satisfy the kinematic constraints. We demonstrate the performance of our algorithm on a 38-DOF articulated human-like model and generate efficient motion strategies for walking, sitting and grabbing objects in complex CAD models.

 

11:35-12:00 Motion Autonomy for Humanoid Robots: research activities in CNRS-AIST JRL

Abderrahmane Kheddar*, Jean-Paul Laumond+, Kazuhito Yokoi*, Eiichi Yoshida*, *CNRS-AIST JRL, UMI3218/CRT France/Japan, +LAAS-CNRS, France

 

Abstract: We present an overview of research activities on humanoid motion planning by on the French-Japanese collaboration: now reinforced by establishing a new international research unit CNRS-AIST JRL, UMI3218/CRT. We present the results in humanoid motion planning for: 3D whole-body collision-free dynamic motion, motions including contacts, whole-body manipulation, and human-humanoid interaction. The research results have been validated by humanoid HRP-2 and shared through common development environments.

 

12:00 Closing

 

Registration:                     Free

 

Contact:

Eiichi Yoshida, Dr., Eng.

E-mail: yoshida (at) laas.fr

Kensuke Harada, Dr., Eng.

E-mail: kensuke.harada (at) aist.go.jp
Intelligent Systems Research Institute
National Institute of Advanced Industrial Science and Technology(AIST)
1-1-1 Umezono, Tsukuba, Ibaraki 305-8568, JAPAN