Development and Implementation of Grasp Algorithm for Humanoid

Robot AR-601M

Kamil Khusnutdinov

1 a

, Artur Sagitov

1 b

, Ayrat Yakupov

1 c

, Roman Meshcheryakov

2 d

Kuo-Hsien Hsia

3 e

, Edgar A. Martinez-Garcia

4 f

and Evgeni Magid

1 g

Department of Intelligent Robotics, Higher Institute for Information Technology and Intelligent Systems,

Kazan Federal University, 35 Kremlyovskaya street, Kazan, Russian Federation

V.A. Trapeznikov Institute of Control Sciences of Russian Academy of Sciences, 65 Profsoyuznaya street, Moscow, 117997,

Russian Federation

Department of Electrical Engineering, Far East University, Zhonghua Road 49, Xinshi District, Tainan City, Taiwan

Universidad Autonoma de Ciudad Juarez, Cd. Juarez Chihuahua, 32310, Mexico

Keywords:

Algorithm, Grasp Planning, Hand Pose Detection, Humanoid Robot.

Abstract:

In robot manipulator control, grasping different types of objects is an important task, but despite being a subject

of many studies, there is still no universal approach. A humanoid robot arm end-effector has a signiﬁcantly

more complicated structure than the one of an industrial manipulator. It complicates a process of object

grasping, but could possibly make it more robust and stable. A success of grasping strongly depends on

a method of determining an object shape and a manipulator grasping procedure. Combining these factors

turns object grasping by a humanoid into an interesting and versatile control problem. This paper presents a

grasping algorithm for AR-601M humanoid arm with mimic joints in the hand that utilizes the simplicity of

an antipodal grasp and satisﬁes force closure condition. The algorithm was tested in Gazebo simulation with

sample objects that were modeled after selected household items.

1 INTRODUCTION

In the ﬁeld of manipulator control a process of grasp-

ing an object is a serious research problem. There is

a growing interest in ﬁnding solutions of this problem

that could be implemented for any humanoid robot.

The area of humanoid robots application is extensive

as they can work in variety of environments, including

factories or social events. A sheer variety of objects

that need to be manipulated by humanoids in different

environments is daunting, considering differences in

objects characteristics, such as shape, size and weight.

In order to successfully grasp an object, a robot

hand must adapt to particular characteristics of the

https://orcid.org/0000-0001-5699-1294

https://orcid.org/0000-0001-8399-460X

https://orcid.org/0000-0001-9977-2833

https://orcid.org/0000-0002-1129-8434

https://orcid.org/0000-0003-1320-673X

https://orcid.org/0000-0001-9163-8285

https://orcid.org/0000-0001-7316-5664

object. Also a robot need to consider environmental

conditions when performing grasping actions. Clut-

ter in a grasping area (Zhu et al., 2014) hinders ob-

ject detection, correct pose estimation and evaluat-

ing characteristics of the object’s surface. In addi-

tion, it is required to consider a possibility that sur-

face characteristics of an object may change under

some conditions. For example, it may become wet

or be deformed due to compression of the object by

robot ﬁngers. In such cases, interaction between the

robot hand and the object surface may change sporad-

ically. These limitations should be considered during

planning of a grasping action. A solution of objects

grasping by a humanoid should include the follow-

ing steps: obtaining information about a target object,

evaluating grasping position(s), and planning move-

ments (possibly, solving inverse kinematics problems

for ﬁngers and hand).

Usually, a grasping scene is represented as a 3D

point cloud, which is further used to extract various

data about a target object (Lippiello et al., 2013).

Evaluating position and orientation of the target ob-

Khusnutdinov, K., Sagitov, A., Yakupov, A., Meshcheryakov, R., Hsia, K., Martinez-Garcia, E. and Magid, E.

Development and Implementation of Grasp Algorithm for Humanoid Robot AR-601M.

DOI: 10.5220/0007921103790386

In Proceedings of the 16th International Conference on Informatics in Control, Automation and Robotics (ICINCO 2019), pages 379-386

ISBN: 978-989-758-380-3

379

ject can help determining a type of grasp routine to

be used (Huang et al., 2013). In addition, perform-

ing shape estimation enables to use predeﬁned rep-

resentation of objects, e.g., in a form of geometric

primitives (Herzog et al., 2014), and then to use pre-

calculated grasp point(s). Using representation of the

target object as a 3D point cloud, simple antipodal

grasps could be generated (Pas et al., 2017). More-

over, if a hand is not able to reach a required grasping

position, the robot could attempt performing a small

pushing/pulling movement in order to relocate an ob-

ject inside the robot hand’s workspace.

Various criteria could be employed for estimating

a grasp quality (Roa and Su

arez, 2015), (Feix et al.,

2013). For example, it could be a degree of ﬁxation

rigidity of an object in a hand (Chalon et al., 2013).

A grasp should be stable and robust, i.e., an object

should not move freely in a grasping hand. Addition-

ally, grasping and manipulation should always con-

sider a possibility of occlusion (Romero et al., 2013).

This paper presents a grasp algorithm for hu-

manoid robot AR-601M (Magid and Sagitov, 2017),

which ﬁngers are constructed with mimic joints. The

algorithm provides an antipodal grasp and satisﬁes

force closure condition using mimic joints. It was

tested in simulation within Gazebo environment us-

ing a 3D model of the right arm of humanoid robot

AR-601M. The tests utilized synthetic objects that

were constructed using their physical prototypes in

real world, which will be further used for the algo-

rithm experimental validation with the real robot.

The rest of the paper is organized as follows. Sec-

tion 2 presents a literature review. Section 3 describes

kinematics of AR-601M humanoid robot arm and its

right arm workspace. Section 4 describes the grasp

algorithm implementation. Section 5 presents the re-

sults of Gazebo simulation using 3D models of AR-

601M right arm and the synthetic objects. Finally, we

conclude and discuss future work in Section 6.

2 LITERATURE REVIEW

This section brieﬂy familiarizes a reader with grasp

planning strategies and approaches, techniques for

their implementation and methods of grasp evalua-

tion.

Selecting an optimal grasp from all possible alter-

natives is a challenging problem and there are a num-

ber of approaches. Empirical approaches, for exam-

ple, are searching for the best grasp in available exper-

imental data utilizing criteria based on target object

features. Authors of data-driven based grasp synthe-

sis review revealed interrelation with analytical meth-

ods (Bohg et al., 2014) and identiﬁed existing grasp-

ing open problems. The main difﬁculty lies in the de-

sign of the appropriate structure representing known

object grasps in terms of robot perception that will fa-

cilitate further search and synthesis. The need for suf-

ﬁciently big prior data to achieve a high success rate is

one of the biggest disadvantages, however, simulation

can be used to generate necessary data.

To solve the problem of generating a stable and ro-

bust grasp, authors in (Lin and Sun, 2015) presented

an approach to a grasp planning that can reconstruct a

simpliﬁed human grasp strategy (represented by grasp

type and thumb positioning) observing human’s ac-

tions in similar manipulation. A learned strategy is a

represention of a recipe to do manipulation with ob-

jects of a particular geometry. The integration of such

strategies into grasp planning procedure acts as a con-

straint on a search space, thus allowing planning to be

computed much faster, still providing sufﬁcient space

not restricting arm agility. A resulting approach inte-

grating a set of learned strategies was compared with

the GraspIt! grasp planner, which doesn’t utilize sim-

ilar constraints. A comparison between approaches

showed that the proposed approach generates grasps

much faster. Generated grasps were similar in con-

ﬁgurations to human operations. Paper didn’t include

tests on novel objects, therefore it is not possible to

determine if the approach can be extended to a new

class of objects.

Hand trajectories, captured during object grasp-

ing tasks by human subjects, was used to deﬁne

and evaluate a set of indicators that were further

used to determine and transfer optimal grasp to

robot hand(Cordella et al., 2014). Based on the

Nelder–Mead simplex method indicators estimates

the optimal grasp conﬁguration for a robotic hand,

considering the limitations that arise when determin-

ing the grasp conﬁguration. Using cross-cylinder as a

target object grasp task was executed by the six hu-

man subjects. The advantages of the proposed al-

gorithm are that it has a reduced computational cost.

With its help it possible to identify and extract quan-

titative indicators to describe the optimal grasp poses

and their reproduction by humanoid robot’s hand. It

can predict the ﬁnal hand position after the movement

and the optimal conﬁguration of the ﬁngers for grasp

execution as soon as it is provided with information

about the size of an object and its location. Its disad-

vantage is that it strongly depends on the similarity of

the robotic hand with the human hand, and is that it

does not consider possible slippage of the hand during

the grasp an object.

The paper (Bullock et al., 2013) presents a classi-

ﬁcation scheme for humanoid arm manipulations.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

380

Taxonomy is deﬁned on various manipulation behav-

iors according to the nature of the contact with exter-

nal objects and the movement of an object. It allows

deﬁning simple criteria that can be applied together in

order to easily classify a wide range of manipulation

behavior for any system in which the hand can be de-

ﬁned. It is argued that the dexterous movements of

the hand can offer an expanded workspace manipula-

tion and improved accuracy with reduced energy con-

sumption but at the cost of added complexity. The ad-

vantage of the proposed classiﬁcation scheme is that

it creates a descriptive structure that can be used to

effectively describe hand movements during manipu-

lation in various contexts and can be combined with

existing object-oriented or other taxonomies to pro-

vide a complete description of a particular manipula-

tion task. Its disadvantage is that during the manipu-

lations it implies the obligatory movement of the hand

and does not consider situations in which the manip-

ulations can be carried out by the movement of the

ﬁngers only, with a ﬁxed hand.

To solve the problem of analyzing the movement

of arms, paper (W

org

otter et al., 2013) presented an

ontology tree for manipulation tasks based on se-

quences of graphs where each graph represents the

relationship between various manipulated objects ex-

pressed in adjacency to each other. The ontology tree

can also be used as a powerful abstraction used in

robotic applications to represent complex manipula-

tion as a set of simple actions (called a chain of se-

mantic events) The advantage of presented ontology

is that it allowed determining about 30 types of fun-

damental manipulations, obtained as a result of an at-

tempt to structure manipulations in space and time. Its

disadvantage lies in the fact that to determine the sim-

ilarity of types of manipulations, an average thresh-

old value of similarity of species equal to 65% was

used, which generally demonstrates not high accuracy

in determining the difference between species.

The article (Daﬂe et al., 2014) proposed 12 pos-

sible types of grasps, with the help of which it is

possible to carry out manipulations under the inﬂu-

ence of external forces. An object is manipulated

through precisely controlled ﬁngertip contacts, con-

sidering non-hand resources. The advantage of the

proposed solution is that it allows carrying out ma-

nipulations in conditions close to real conditions, in

comparison with situations in which the manipulation

is carried out with static hands. And in that it allows

looking at the process of making manipulations from

a different point of view, going beyond the tradition-

ally considered clever manipulations. The disadvan-

tage is that among the presented types of grasp there

are no grasps identical to human grasps.

During the generation of grasps, there can en-

counter with a problem of determining a grasp strat-

egy that is capable of ensuring the compatibility of

an object deﬁnition tasks and the implementation of

its grasp, as well as capable of ensuring adaptability

to new objects. To solve this problem, an investiga-

tion (Sahbani et al., 2012) of analytical and empiri-

cal approaches to the construction of a grasp strategy

presents a review of algorithms for synthesizing the

grasp of three-dimensional objects. The advantage of

the article is that its authors managed to ﬁnd a pos-

sible problems solution that consists in introducing

into the work of the robot the ability to autonomously

identify the signs of a new object, with the help of

which it can understand what object is in front of it.

Its disadvantages include the absence of various sim-

ulations of grasp using information about the cases in

which each of the approaches is most applicable.

An approach (Pham et al., 2015) for evaluating

contact forces based only on visual input data pro-

vided by a single RGB-D camera aims to solve the

problem of estimating forces applied by hand to an

object. The input information is extracted using vi-

sual tracking of the hand and an object to assess their

position during the manipulation. After that, the kine-

matics of hand movement is calculated using a new

class of numerical differentiation operators. Further

the estimated kinematics is fed into the program that

returns the desired result: the minimum distribution

of force along with an explanation of the observed

movement. The advantage of the proposed approach

is the ability of solving the problem of determining

the contact points of a hand and an object when strong

occlusions occurred, using an approach based on the

assumption that the closed ﬁngers remain in their last

observed position until they are visible again. Its dis-

advantage is that it cannot solve the problem of de-

termining contact points when performing clever ma-

nipulation, with moving a ﬁnger or sliding, due to the

fact that it uses the above presented assumption that

is not fair in the case of dexterous manipulations.

In a multitude of manipulation scenarios, a robot

may come into collision with objects of unknown

shape: as they rotate, they will remain symmetrical.

For such objects, there is no known three-dimensional

model. The problem arises of assessing the 3D pos-

ture and the shape of such objects, which prevents one

from understanding what this object is. To solve this

problem, the paper (Phillips et al., 2015) proposes

an algorithm for the simultaneous evaluation of the

posture and shape of an object without using cross-

sections. The solution uses the properties of the pro-

jective geometry of the surface of revolution. It re-

stores the three-dimensional pose and shape of an ob-

Development and Implementation of Grasp Algorithm for Humanoid Robot AR-601M

381

ject with an unknown surface of rotation from two

points of view: suitable types of known, relative ori-

entation. The advantages of the proposed algorithm

are that it can work even when only one of the two

visible contours of the surface of revolution and that

it is suitable for determining the posture and shape of

transparent objects, providing clear contours of such

objects. Information about whether the algorithm can

estimate the position of an object of a similar object

and restore its shape from noisy images is not pro-

vided.

During the manipulation, one potential problem

is the arrangement of feedback acquisition from the

hand. To solve this problem, the authors of the paper

(Cai et al., 2016) presented the hypothesis that for ac-

curate recognition of manipulations, it is necessary to

model the types of hands and attributes of objects be-

ing manipulated. The paper presents a uniﬁed model

for evaluating the manipulation of the hand and an ob-

ject, in which the observation of the manipulation is

performed from a wearable camera. From the areas

of the hand detected on one image, the type of grasp

is recognized, and its attributes are determined from

the detected parts of an object. The nature of the ma-

nipulation is determined by the relationship between

the type of grasp and the attributes of an object, rep-

resenting a set of beliefs about them embedded in the

model. The paper provides a model estimate for the

correlation between the type of grasp and shape of an

object. The advantage of the model is that it exceeds

the traditional model that does not consider the inter-

relation of such semantic constraints as the type of

grasp and the attributes of an object. Its disadvantage

is that it has an average recognition accuracy of the

type of grasp at 61.2%.

3 AR-601M ARM KINEMATICS

Each arm of AR-601M humanoid robot has 20 de-

grees of freedom (DOF), where 7 DOFs correspond

to the arm and 13 DOFs correspond to the ﬁve ﬁn-

gers of an arm. The ﬁngers are designed with mimic

joints in all phalanges but proximal. The 3D robot

model is constructed in a Gazebo simulator environ-

ment (Shimchik et al., 2016). Figure 1 demonstrates

a 3D model of AR-601M humanoid right arm, which

is an exact replica of the real robot right arm. The

robot model was integrated into the Robot Operating

system (ROS) and MoveIt! motion planner frame-

work. RRTConnect algorithm was chosen for tra-

jectories planning in control of AR-601M arm move-

ments (Lavrenov and Zakiev, 2017).

The arm movement planning requires solving in-

Figure 1: A 3D model of AR-601M humanoid right arm.

verse kinematics (IK) problem ﬁrst. Since ROS con-

tains a set of IK solvers, the solution of IK problem

was reduced to a suitable plugin selection that suited

our robot’s arm constraints and desired characteris-

tics. We selected k dl kinematics plugin, which is an

effective tool for solving IK for manipulators with 6

or more DOFs. Figure 2 shows a pre-grasp position

that was calculated with kdl kinematics plugin.

Figure 2: Pre-grasp position of AR-601M arm.

To apply the 3D model of AR-601M humanoid

right arm it is necessary to calculate its workspace

in order to prepare a scene and calculate the arm

movements. The workspace was calculated numeri-

cally through applying series of forward kinematics

cases until the representative density of workspace

reachable (by the end effector) points was achieved.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

382

We used Matlab environment and Robotics Tool-

box (Corke, 2017) to calculate reachable workspace

of AR-601M right arm (Fig. 3).

Figure 3: Reachable workspace of AR-601M right arm that

was calculated in Matlab environment.

4 GRASPING ALGORITHM

Because of designing a humanoid hand as a set of ﬁn-

gers, some grasp algorithms for humanoids include

calculation of contact points (Yu et al., 2017), (V Le

et al., 2010). For our algorithm we assume that it

is possible to execute a reliable grasp using simple

approach without calculating contact points, since it

is enough to apply basic forces to an object surface

within the plane (similarly to antipodal grasp execu-

tion). At this stage, an object may not be rigidly ﬁxed

in the hand yet, but it will not fall out of the grasp.

Next, the robot completes the grasp applying addi-

tional forces and performing ﬂexion of the remain-

ing ﬁngers. The advantage of the algorithm is that

it avoids contact points calculation. Its disadvantage

is that depending on a generated handle location, the

ﬂexion of the ring and the pinkie ﬁngers could be ex-

ecute outside of the object. Therefore, it is not always

possible to satisfy the force closure condition.

4.1 Obtaining Object Data from a Point

Cloud

Before grasping an object it is necessary to ﬁnd out in-

formation that will allow the grasping. To obtain such

information various tools could be used, but most of

these tools employ a 3D point cloud as a source for

further data extraction. The idea behind is that ob-

jects and a scene of manipulation are represented as a

combination of a large number of 3D points that are

located very close to each other. A point cloud could

be obtained with RGBD cameras, image depth sen-

sors or motion sensors, which are used as image depth

sensors. They allow receiving both a color image and

its 3D representation in a form of a 3D point cloud.

Figure 4 shows a point cloud for 3D models of ob-

jects (a bottle and a ladle) that were obtained with

Microsoft Kinect sensor within Gazebo simulator.

Figure 4: 3D models of bottle (top) and ladle (bottom) rep-

resented as 3D point clouds.

4.2 Objects Description and Grasp

Geometry

Various tools that allow extracting data (which is nec-

essary in order to grasp an object) from a 3D point

cloud, could provide data about contact points and

their coordinates in 3D space, an area from which

an object should be extracted, and objects geome-

try. Next, the extracted data is used for grasp exe-

cuting. Some tools can generate grasps, but typically

these are simple antipodal grasps, which are carried

out by a two-ﬁnger gripper with the ﬁngers moving in

the same plane and driven by a prismatic joint. ROS

package handle detector (Pas and Platt, 2014) is a

tool that extracts data about an object from a scene by

analyzing objects surface (represented as a 3D point

cloud), identiﬁes and visualizes an area of the object

(referred as a handle) that should be grasped. Each

handle is represented with a set of cylinders with their

own parameters, providing handle-related data that

includes handle position, orientation, radius and ex-

Development and Implementation of Grasp Algorithm for Humanoid Robot AR-601M

383

tent. Figure 5 shows handles for 3D models of a bottle

and a ladle that were generated using handle detector

tool.

Figure 5: Handles for 3D models of a bottle (top) and a

ladle (bottom) are shown with cyan color.

4.3 Calculating a Target Pose of an

Object Grasping Hand

Before grasping an object it is necessary to ﬁnd a tar-

get point for a hand that it needs to reach in order to

perform the antipodal grasp. When the hand reaches

its target position, an object should lie between the

thumb and the other ﬁngers. The index ﬁnger was se-

lected as a pair for the thumb in the antipodal grasp.

With handle detector tool this corresponds to a situ-

ation where the hand reaches its target point, and thus

handle center coordinates (point A) should align with

point B. To implement this idea, the coordinates of

thumb tip (point C) and index tip (point D) in their de-

fault positions (an open antipodal gripper) are deter-

mined. Next, coordinates of point B are determined

and subtracted from coordinates of point A in order

to calculate a displacement vector. To determine co-

ordinates of the hand target point, the displacement

vector coordinates are added to the coordinates of the

current hand position. Figure 6 visualizes these an-

tipodal grasp geometry calculation.

Figure 6: Components for calculation a target point for a

hand. A is a center of a handle, B is a point that is equidistant

from tips of a thumb and an index ﬁngers, C is a tip of the

thumb, D is a tip of the index ﬁnger.

4.4 Determining Positions of Fingers to

Grasp an Object

Each ﬁnger tip position is determined with calcula-

tions of ﬁnger joints’ rotation angles. Since the joints

are mimic, this reduces to calculating rotation angles

for active joint of a ﬁnger, which is a ﬁrst joint. Us-

ing handle detector tool we obtain diameter d of a

handle’s cylinder. Next, thumb and index ﬁngers ﬂex-

ing is simulated and distance S between their tips

is tracked and compared to d, resulting into optimal

mimic joint rotation angles selection that minimizes

the difference between d and S. To calculate mimic

joint rotation angles of other ﬁngers, it is necessary to

consistently increase the angles of their active joints

and select the ones that help satisfying the force clo-

sure condition. Figure 7 visualizes these calculations.

Figure 7: Calculating rotation angles: d is a handle diame-

ter, S is a distance between thumb and index tips.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

384

5 SIMULATION OF GRASPING

AN OBJECT

The grasp algorithm for the AR-601M humanoid us-

ing handle detector tool consists of the following se-

quence of actions:

1. Estimate coordinates of a handle’s center;

2. Use coordinates of ﬁngertips that are in an open

state to calculate a point that is equidistant from

the ﬁngertips;

3. Subtract coordinates of the point founded in Step

2 from the handle’s center coordinates, ﬁnd the

hand offset vector coordinates;

4. Add the coordinates of Step 3 to current hand po-

sition coordinates, ﬁnd coordinates of a new hand

target position that is necessary to grasp an object;

5. Estimate trajectory of grasp ﬁngers’ ﬂexion to cal-

culate distances between ﬁngertips and the han-

dle’s circumference. Active joints’ angles that are

minimizing the difference between the distance

between the ﬁngertips and the handle circumfer-

ence are the required angles of ﬂexion required to

grasp the object.

The algorithm was veriﬁed in virtual experiments

that were performed in Gazebo simulation. Five dif-

ferent models of real world objects were tested: a bot-

tle, a rectangular box of vitamins, a juice box, a ladle

and a plastic cup without a handle. Figure 8 shows

virtual experiments of grasping 3D models of a bottle

and a ladle by AR-601M humanoid robot using the

proposed grasping algorithm.

6 CONCLUSION AND FUTURE

WORK

We presented the development and implementation of

a grasping algorithm for AR-601M humanoid robot

that utilizes a simplicity of an antipodal grasp and

satisﬁes force closure condition using mimic joints.

The algorithm was tested in Gazebo simulation en-

vironment with ﬁve different synthetic objects that

were constructed using their physical prototypes in

real world. Our approach’s advantages and disadvan-

tages were discussed. As a part of our future work

we plan to validate the algorithm in real world envi-

ronment with AR-601M humanoid for execution of

pick and place operations. RGBD sensor will be used

to provide the 3D point cloud data for the algorithm.

One of the particular tasks that will be implemented

using the proposed algorithm is door handle grasp-

ing and door opening. This is a necessary task of

(a) (b)

(c)

Figure 8: Grasping objects with the proposed algorithm (a),

(b) a bottle; (c) a ladle.

AR-601M humanoid in victim search mission within

household environment that is a part of our large-scale

project in robotized urban search and rescue.

ACKNOWLEDGEMENTS

The reported study was funded by the Russian Foun-

dation for Basic Research (RFBR) according to the

research project No. 19-58-70002. Part of the work

was performed according to the Russian Government

Program of Competitive Growth of Kazan Federal

University.

REFERENCES

Bohg, J., Morales, A., Asfour, T., and Kragic, D. (2014).

Data-driven grasp synthesis—a survey. IEEE Trans-

actions on Robotics, 30(2):289–309.

Bullock, I. M., Ma, R. R., and Dollar, A. M. (2013). A

hand-centric classiﬁcation of human and robot dex-

terous manipulation. IEEE Transactions on Haptics,

6(2):129–144.

Cai, M., Kitani, K., and Sato, Y. (2016). Understanding

hand-object manipulation with grasp types and object

Development and Implementation of Grasp Algorithm for Humanoid Robot AR-601M

385

attributes. In Robotics conference: Science and Sys-

tems 2016.

Chalon, M., Reinecke, J., and Pfanne, M. (2013). Online in-

hand object localization. In 2013 IEEE/RSJ Interna-

tional Conference on Intelligent Robots and Systems,

pages 2977–2984.

Cordella, F., Zollo, L., Salerno, A., Accoto, D.,

Guglielmelli, E., and Siciliano, B. (2014). Human

hand motion analysis and synthesis of optimal power

grasps for a robotic hand. International Journal of

Advanced Robotic Systems, 11(3):37.

Corke, P. I. (2017). Robotics, Vision & Control: Fundamen-

tal Algorithms in MATLAB. Springer, second edition.

ISBN 978-3-319-54412-0.

Daﬂe, N. C., Rodriguez, A., Paolini, R., Tang, B., Srini-

vasa, S. S., Erdmann, M., Mason, M. T., Lundberg, I.,

Staab, H., and Fuhlbrigge, T. (2014). Extrinsic dex-

terity: In-hand manipulation with external forces. In

2014 IEEE International Conference on Robotics and

Automation (ICRA), pages 1578–1585.

Feix, T., Romero, J., Ek, C. H., Schmiedmayer, H., and

Kragic, D. (2013). A metric for comparing the anthro-

pomorphic motion capability of artiﬁcial hands. IEEE

Transactions on Robotics, 29(1):82–93.

Herzog, A., Pastor, P., Kalakrishnan, M., Righetti, L., Bohg,

J., Asfour, T., and Schaal, S. (2014). Learning of grasp

selection based on shape-templates. Auton. Robots,

36:51–65.

Huang, B., El-Khoury, S., Li, M., Bryson, J. J., and Billard,

A. (2013). Learning a real time grasping strategy. In

2013 IEEE International Conference on Robotics and

Automation, pages 593–600.

Lavrenov, R. and Zakiev, A. (2017). Tool for 3d gazebo map

construction from arbitrary images and laser scans.

In 2017 10th International Conference on Develop-

ments in eSystems Engineering (DeSE), pages 256–

261. IEEE.

Lin, Y. and Sun, Y. (2015). Robot grasp planning based

on demonstrated grasp strategies. The International

Journal of Robotics Research, 34:26–42.

Lippiello, V., Ruggiero, F., Siciliano, B., and Villani, L.

(2013). Visual grasp planning for unknown objects us-

ing a multiﬁngered robotic hand. IEEE/ASME Trans-

actions on Mechatronics, 18(3):1050–1059.

Magid, E. and Sagitov, A. (2017). Towards robot fall de-

tection and management for russian humanoid ar-601.

In KES International Symposium on Agent and Multi-

Agent Systems: Technologies and Applications, pages

200–209. Springer.

Pas, A., Gualtieri, M., Saenko, K., and Platt, R. (2017).

Grasp pose detection in point clouds. The Interna-

tional Journal of Robotics Research, 36(13-14):1455–

1473.

Pas, A. and Platt, R. (2014). Localizing handle-like grasp

affordances in 3d point clouds. In The 14th Interna-

tional Symposium on Experimental Robotics, volume

109.

Pham, T.-H., Kheddar, A., Qammaz, A., and Argyros, A. A.

(2015). Towards force sensing from vision: Observing

hand-object interactions to infer manipulation forces.

In 2015 IEEE Conference on Computer Vision and

Pattern Recognition (CVPR), pages 2810–2819.

Phillips, C. J., Lecce, M., Davis, C., and Daniilidis, K.

(2015). Grasping surfaces of revolution: Simultane-

ous pose and shape recovery from two views. In 2015

IEEE International Conference on Robotics and Au-

tomation (ICRA), pages 1352–1359.

Roa, M. A. and Su

arez, R. (2015). Grasp quality mea-

sures: review and performance. Autonomous Robots,

38(1):65–88.

Romero, J., Kjellstr

om, H., Ek, C. H., and Kragic, D.

(2013). Non-parametric hand pose estimation with

object context. Image and Vision Computing, 31:555–

564.

Sahbani, A., El-Khoury, S., and Bidaud, P. (2012). An

overview of 3d object grasp synthesis algorithms.

Robotics and Autonomous Systems, 60:326–336.

Shimchik, I., Sagitov, A., Afanasyev, I., Matsuno, F., and

Magid, E. (2016). Golf cart prototype development

and navigation simulation using ros and gazebo. In

MATEC Web of Conferences, volume 75, page 09005.

EDP Sciences.

V Le, Q., Kamm, D., F Kara, A., and Y Ng, A. (2010).

Learning to grasp objects with multiple contact points.

In 2010 IEEE International Conference on Robotics

and Automation (ICRA), pages 5062 – 5069.

org

otter, F., Aksoy, E. E., Kr

uger, N., Piater, J., Ude, A.,

and Tamosiunaite, M. (2013). A simple ontology of

manipulation actions based on hand-object relations.

IEEE Transactions on Autonomous Mental Develop-

ment, 5(2):117–134.

Yu, D., Yu, Z., Zhou, Q., Chen, X., Zhong, J., Zhang, W.,

Qin, M., Zhu, M., Ming, A., and Huang, Q. (2017).

Grasp optimization with constraint of contact points

number for a humanoid hand. In 2017 IEEE Interna-

tional Conference on Robotics and Biomimetics (RO-

BIO), pages 2205–2211.

Zhu, M., G. Derpanis, K., Yang, Y., Brahmbhatt, S., Zhang,

M., Phillips, C., Lecce, M., and Daniilidis, K. (2014).

Single image 3d object detection and pose estima-

tion for grasping. In 2014 IEEE International Con-

ference on Robotics and Automation (ICRA), pages

3936–3943.

ICINCO 2019 - 16th International Conference on Informatics in Control, Automation and Robotics

386