Optimization-based Posture Generation for Whole-body Contact Motion by Contact Point Search on the Body Surface
Masaki Murooka, Kei Okada, Masayuki Inaba · 2025
Read Optimization-based Posture Generation for Whole-body Contact Motion by Contact Point Search on the Body Surface by Masaki Murooka, Kei Okada, Masayuki Inaba (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2505-23501.txt.
Introduction
Masaki Murooka, Kei Okada, and Masayuki Inaba present a method that plans robot poses using contact across the whole body, in a paper from 2025. Robots that lean, brace, or hug objects with more than their hands gain steadiness and strength. Planning such poses by hand is slow. Their method lets a program search for good poses on its own.
The trick is to mark each contact point with two values that place it on the robot's skin. A solver then adjusts the joints and the contact points at the same time. It climbs toward poses that meet the task and stay balanced. In trials, robots handled objects well by pressing them with arms and torso, not just grippers. The approach opens a wider range of moves for humanoid and other multi-joint robots.
Catalog record
| CREATOR | Masaki Murooka, Kei Okada, Masayuki Inaba |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 38 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2505-23501
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Optimization-based Posture Generation for Whole-body Contact Motion by Contact Point Search on the Body Surface Masaki Murooka, Kei Okada, Masayuki Inaba Abstract Whole-body contact is an effective strategy for improving the stability and efficiency of the motion of robots. For robots to automatically perform such motions, we propose a posture generation method that employs all available surfaces of the robot links. By representing the contact point on the body surface by two-dimensional configuration variables, the joint positions and contact points are simultaneously determined through a gradient-based optimization. By generating motions with the proposed method, we present experiments in which robots manipulate objects effectively utilizing whole-body contact. I Introduction Robots must perform dexterous and high-load operations autonomously to expand the range of applications to daily life support and disaster response. In the robot motions performed so far, the robots usually contact with the environment only through certain body points (e.g., gripper and foot), making the motions unstable and inefficient. In this study, to solve such a problem, we propose a method to generate motion through contact at any point on the robot’s body surface. This is a versatile method in which the contact points on the body surface are expressed as configuration variables and searched together with joint positions by the conventional calculation procedure of inverse kinematics. We show the effectiveness of the proposed method by applying to a variety of whole-body contact motions. I-A Related Works I-A 1 Whole-body Contact Motion by Robot Whole-body contact motions, that effectively use the robot’s body, have been widely studied. Whole-body manipulation [ 1 ] and getting up [ 2 ] are typical motions and realized by real robots. To generate these motions, the key postures are taught online [ 3 ] or the contact points are predetermined manually [ 4 ] . Many of these approaches follow human intervention or ad hoc rules, and the general method for automatically generating whole-body contact motion as proposed in this study is yet to be established. The proposed method is also effective in utilizing the whole-body contacts in the multi-contact motion of humanoid robots, which has been widely studied in recent years [ 5 ] [ 6 ] . I-A 2 Grasp Planning of Multi-Fingered Hand Power grasp [ 7 ] with a multi-fingered hand belongs to the same kind of motion because the contact points are selected from the finger surface. In many of the grasp planning methods, the collision points are searched in the simulator by closing the fingers from the sampled initial posture [ 8 ] [ 9 ] . Although this method is applicable for generating whole-body contact motion [ 1 ] , the sampling-based method has disadvantages including high computational costs due to repeated trials and difficulty in generating time-series postures or postures with good evaluation values (e.g., joint torque and nominal posture). In this study, we employ the relatively efficient gradient-based optimization. I-A 3 Optimization-based Motion Generation Recently, the optimization-based motion generation has been actively studied [ 10 ] . Inverse kinematics (IK) is the fundamental but non-trivial problem in motion generation [ 11 ] . Our method is positioned as an extension of inverse kinematics because the conventional point-to-point IK is generalized to a body-to-body IK. To apply the gradient-based optimization that is widely used in motion generation, we introduce configuration variables representing points on the body surface and derive the gradient with respect to the variables. Another approach has been proposed in which optimization problems are formulated on non-Euclidean manifolds to handle the surface contacts [ 12 ] . I-A 4 Geometric Model Representation The problem with searching a point on the body surface by the gradient-based optimization is the difficulty in calculating the gradient of the point on the edge of the robot link represented by a polyhedron. A solution to this problem is approximating the shape of the robot link to a smooth shape represented by a sphere-torus-patch [ 13 ] or Catmull-Clark subdivision surface [ 14 ] . However, implementation and calculation costs are involved to approximate the robot model. In this study, normal direction smoothing enables calculation of the gradient from the raw convex mesh of the robot model without conversion. I-B Contributions of this Study The contributions of this study are as follows: (i) an optimization problem that simultaneously searches joint positions and the contact points on the body surface is formulated, (ii) a smoothed normal direction on the body surface is introduced for gradient-based optimization, (iii) various whole-body contact motions are shown through simulation and real robots. In the following, Sec. II describes the problem and provides an outline of the solution, and Sec. III describes contact point searching on the body surface in detail. Numerical examples are given in Sec. IV, and Sec. V demonstrates applications of the whole-body contact motion. II Overview of Whole-body Contact Motion Generation II-A Problem Definition The simplest problem setting for whole-body contact motion generation is provided. The task goals are defined from the following information: • Kinematics model of the robot, object, and environment. • Pairs of bodies to be contacted at time index t , \langle\mbox{\boldmath{$B$}}_{t,n,1},\mbox{\boldmath{$B$}}_{t,n,2}\rangle\ (t=1,\cdots,T,\ n=1,\cdots,N(t)) . where T is the number of sampled time, and N(t) is the number of contact pairs at time index t . The kinematics model must contain information of the links and joints. The body represents a robot link, object, or environment that is modeled as a plane or convex polyhedron. Examples of the body are shown in Fig. 1 , and the problem task is defined through the following: find { joint position, contact point } t=1,⋯,T (1) s.t. \mbox{\boldmath{$B$}}_{t,n,1} and \mbox{\boldmath{$B$}}_{t,n,2} contact. (t=1,\cdots,T,\ n=1,\cdots,N(t)) The following constraints are satisfied: joint position range and collision avoidance. This problem is termed posture generation when T=1 , and motion generation when T\geq 2 . An example of the problem of posture generation is displayed in Fig. 2 (A). robot link object ground Fig. 1: Examples of body to be contact. The body is modeled as a plane or convex polyhedron. For robot links, we use the convex polyhedron generated from the detailed mesh model automatically. Although the number of vertices in the body is high (about 50), the proposed optimization method based on normal smoothing works properly. We use simple primitive shapes (cylinder, cube, and sphere.) for the object and environment. (A) posture generation (B) contact Fig. 2: Example of posture generation. (A) The figure shows an example of the whole-body contact posture generated by the proposed method. The robot is in contact with the environment in three places. The contacts of knee and elbow are designated as “link-to-plane” target, whereas the foot contact is designated as a conventional “point-to-point” target. (B) The figure also displays the enlarged view of contact at the knee. The contact positions of both bodies coincide, with reversed contact normal directions. II-B Problem Representation by Optimization The problem ( 1 ) is formulated as an optimization problem as follows: \displaystyle\min_{\mbox{\boldmath{$\hat{q}$}}}\frac{1}{2}\|\mbox{\boldmath{$\hat{e}$}}(\mbox{\boldmath{$\hat{q}$}})\|^{2} (2) \displaystyle{\rm s.t.}\ \ \mbox{\boldmath{$c$}}(\mbox{\boldmath{$\hat{q}$}})\geq\mbox{\boldmath{$0$}} (3) where \mbox{\boldmath{$c$}}(\mbox{\boldmath{$\hat{q}$}}) represents the conditions for joint position range and collision avoidance, while \mbox{\boldmath{$\hat{e}$}}(\mbox{\boldmath{$\hat{q}$}}) represents the conditions for contacts. The configuration \hat{q} is a time series of the joint position \mbox{\boldmath{$\theta$}}_{t} and contact point \mbox{\boldmath{$u$}}_{t} and