HERMES - A Humanoid Mobile Manipulator for Service Tasks

Page created by Jack Molina
 
CONTINUE READING
FSR’97 International Conference on Field and Service Robots. Canberra, December 1997.

                HERMES – A Humanoid Mobile Manipulator for Service Tasks

                                                      Rainer Bischoff

                                    Federal Armed Forces University Munich
                                       Institute of Measurement Science
                             Werner-Heisenberg-Weg 39, 85577 Neubiberg, Germany
                                 E-Mail: Rainer.Bischoff@unibw-muenchen.de

                         Abstract                                 HERMES – Humanoid Experimental Robot for Mobile Ma-
                                                                  nipulation and Exploration Services – are presented.
     To carry out automated service tasks at different
     places within large working environments, a robot            1.1 Advantages of a Humanoid Robot Design
     system able to navigate autonomously and to exe-
                                                                  There are several reasons to take a human as an example for
     cute commanded manipulation tasks is needed. In
                                                                  the design of service robots. The best argument is that such a
     this paper the design concept and the realization
                                                                  robot shall perform its tasks in environments where humans
     of a novel service robot named HERMES are de-
                                                                  work and live, e.g., in apartments, offices, laboratories, res-
     scribed. Its human-like appearance, highly inte-
                                                                  taurants, or hospitals. These environments are designed to
     grated sensors and actuators together with its
                                                                  meet special human characteristics and needs: the space a
     practical task-related intelligence enable the robot
                                                                  human requires (e.g., the width of passageways and doors),
     to perform various service tasks.
                                                                  his working height (e.g., the height of tables or door knobs),
                                                                  his vision height (e.g., the height of door plates) and his
1 Introduction                                                    strength that is needed to manipulate objects (e.g., to open
                                                                  doors). If a robot is placed in such surroundings, it is to be
Service robots that have to operate in many different and
                                                                  designed according to an anthropomorphic model and should
unstructured environments will be of great technological and
                                                                  have comparable sensory and motor skills.
economical importance in the near future [Schraft et al.
                                                                      Another important reason is that service robots have to
1994]. An important characteristic of such robots is that they
                                                                  interact and to communicate with humans by different
have to work in environments that are co-habited by humans
                                                                  means, from touch and gestures to speech. If a robot has a
who are not specially trained to co-operate with robots and
                                                                  humanoid form and exhibits human-like behavior, humans
who are not necessarily interested in them. As a consequence
                                                                  are able to interact in a more natural way [Brooks 1996].
future service robots will need a high degree of robustness,
                                                                      Also, humanoid size and shape can be advantageous for
adaptability and advanced communication abilities in order
                                                                  the representation of knowledge of the environment, because
to deal with unexpected situations. Such robots do not yet
                                                                  the robot can develop human-like sorts of representations
exist.
                                                                  [Johnson 1987].
    To study the possibilities of realizing such robots we have
designed and constructed an experimental robot (Figure 1)
that possesses many of the characteristics that future service
                                                                  1.2 Requirements for the robot HERMES
robots are expected to have. Its flexibility, modularity and      The combination of autonomous navigation with skillful
extendability guarantee that diverse solutions to existing        manipulation is a basic prerequisite for many service robots.
problems can be developed and validated in real-world ex-         Therefore, HERMES should be able to explore unknown
periments. In the sequel the necessity for an anthropomor-        environments in order to navigate therein and to manipulate
phic design is justified and our concept and the realization of   diverse objects according to the given task. It should execute
FSR, Canberra, December 1997                                -2-               Bischoff: Humanoid Mobile Manipulator HERMES

commands like “Take object O from                                                             bile manipulators. [Cameron et al.
the shelf in room 1, transport it to                                                          1993] developed a mobile manipula-
room 2 and place it there on the                                                              tor that moved its arm in a favorable
table!”, without any special modifica-                                                        manipulation position during the
tions of the environment. Vision                                                              docking phase by reactive control
should be the main sensing modality,                                                          methods.
because it has proven to be most pow-
erful in such applications, not only                                                          2.2 Humanoid Robots
with robots but also in nature [Graefe
                                                                                              The transition from mobile manipu-
1992]. The system architecture
                                                                                              lators to humanoid robots is not
should neither rely on exact world
                                                                                              clearly defined in the existing litera-
models nor on an exact knowledge of
                                                                                              ture. In the following a robot is called
optical, kinematical and dynamical
                                                                                              humanoid if it resembles – at least to
parameters. This leads to a high de-
                                                                                              some extend – a human in height,
gree of robustness that will allow
                                                                                              shape, configuration of its degrees of
autonomous operation of service ro-
                                                                                              freedom and both type and arrange-
bots in households and public areas
                                                                                              ment of its main sensors.
as well as in industrial environments.
                                                                                                  [Bergener et al. 1997] have built a
                                                                                              robot that consists of a 7 DOF arm
2 State of the Art                                                                            attached to TRC Labmate platform.
                                                                                              The robot head carries four cameras
Many laboratories work on the devel-                                                          with different focal lengths (two by
opment of components of autono-                                                               two) for navigation and manipulation
mous mobile systems. In most cases                                                            tasks.
research is focused on autonomy or Figure 1: Humanoid robot HERMES with omni-                     Several robots exist, mainly in the
mobility or manipulation. The inte- directional base, two arms, bendable body and two USA and in Japan, that have a more
gration of all three aspects into func- cameras on a pan-tilt platform;                       or less humanoid appearance; how-
tioning systems, mobile manipulators dimensions: 70 cm x 70 cm x 170 cm                       ever, it is not possible to describe
and humanoid robots, came only re-                                                            them all here. Although most of those
cently into the center of interest. In the sequel we give an         systems are immobile, they possess several DOF in bodies,
overview of lately developed systems. Thereafter, project-           manipulators and sensor heads, e.g. [Brooks, Stein 1993],
relevant work of our institute is described.                         [Konno et al. 1997].
                                                                         The most sophisticated humanoids up to date are the
2.1 Mobile Manipulators                                              walking robots P2 and P3 of Honda Motor Corporation. They
                                                                     resemble closely a human in height, shape, configuration of
A mobile manipulator consists of a mobile platform with one          its degrees of freedom. They are able to balance themselves
or more manipulators attached to it. The mobility achieved           automatically if pushed, keep themselves upright according
by the platform yields a significantly enlarged work space           to the angle of a slope, and climb stairs or slopes. These
compared to a fixed manipulator. However, the degrees of             characteristics enable the robots to perform service tasks such
freedom (DOF) of the system are increased and the control of         as pushing a cart and tightening bolts [Honda 1997].
the manipulator becomes more complex. All proposed solu-
tions for this problem have in common that they need te-             2.3 Project-Relevant Work at the Institute of Mea-
diously calibrated sensors and actuators, world models and           surement Science
knowledge of the kinematic structures. Examples of impres-
                                                                     Previous research work conducted at the Institute of Mea-
sive robots that have been realized based on these principles
                                                                     surement Science with the mobile robots ATHENE I and
are, e.g., the assembly robot KAMRO [Lueth et al. 1995], the
                                                                     ATHENE II and the manipulator “Mitsubishi Movemaster”
service robot ROMAN with a remarkable number of key
                                                                     have been key to the design of HERMES.
components for service robots [Daxwanger et al. 1996] and
the rehabilitation robot MOVAID [Dario et al. 1995].                 Mobility
    [Yamamoto 1994] and [Khatib et al. 1995] have devel-             ATHENE II (Figure 2) is a three wheeled vehicle with a
oped force control algorithms for multiple cooperating mo-           monochrome video camera on a one-axis platform and a PC
FSR, Canberra, December 1997                              -3-              Bischoff: Humanoid Mobile Manipulator HERMES

                                                                                                   camera
                                                                  camera                                              gripper

                                                                                                                       object

                                                                  Figure 3: manipulator “Mitsubishi Movemaster” with 5 DOF, two
                                                                  finger gripper and stereo vision system

Figure 2: mobile robot ATHENE II with one-axis camera platform;   mediate test movements. Meanwhile machine learning algo-
dimensions: 135 cm x 70 cm x 110 cm                               rithms have been employed, reducing the grasping time from
for both robot control and as host of a transputer frame grab-    ca. 50 s to 10 s [Xie et al 1997].
ber used for image processing. On this robot a situation-ori-
ented, behavior-based approach for control and navigation         3 A New Design – Concepts and Requirements
has been developed and implemented [Wershofen 1996].
    ATHENE II is able to navigate efficiently in a network of     Some weaknesses and limitations of our institutes’ robots
corridors and open areas, i.e., drive to named locations and      (e.g. insufficient maneuverability, too few degrees of freedom
execute directly specified sequences of behavior patterns. The    and too little payload of the arm, a heterogeneous hard to
robot is able to acquire the necessary knowledge about the        extend overall structure) made a complete redesign neces-
characteristics of the environment by supervised learning.        sary. To open a wide field of possible experiments we have
3-D or 4-D world models are not necessary for neither situa-      decided to realize an humanoid concept.
tion recognition nor execution of behavior patterns [Bischoff         A strictly modular design where all modules have stan-
et al. 1996].                                                     dardized, homogeneous mechanical and electrical interfaces
                                                                  has been considered most important. If these modules are
Manipulation
                                                                  connected via powerful communication links they can be
In general, an exact calibration of the optical parameters and
                                                                  nearly arbitrarily configured and adapted to changing re-
the kinematics of the robot arm is needed for visually guided
                                                                  quirements. This concept of modularity has been pursued
manipulation. This calibration is rather cumbersome and
                                                                  both for the construction of the robot body and for the struc-
needs continuous verification, e.g., after any maintenance.
                                                                  ture of the information processing system. This permits, on
To avoid these limitations, a new method was developed. It
                                                                  the one side, to increase the degrees of freedom of the overall
has led to the realization of a calibration-free manipulator
                                                                  system and, on the other side, to adapt the processing power
with 5 DOF, a two-finger gripper and a stereo vision system
                                                                  by adding computational nodes if this should become neces-
(Figure 3). Flat and elongated cylindrical objects can be lo-
                                                                  sary. Using established components a new robot can thus be
calized and grasped without any knowledge of kinematical or
                                                                  created that is homogenous, flexible, easy to maintain and,
optical parameters; even arbitrary changes of the optical
                                                                  most important, that can be controlled in all of its degrees of
system during manipulation are tolerated [Graefe, Ta 1995].
                                                                  freedom in a uniform manner.
    Basis for this extraordinary robustness is the absence of a
world coordinate system and a direct transition from image
                                                                  3.1 Mobile Base
data to motor commands (control words), without using any
inverse perspective or kinematic transformations. In contrast     Many service robots perform their tasks in environments
to other approaches (e.g. neural networks) no training is         where humans can walk around without any problems. Most
required.                                                         of those environments have in common that they can be
    The first implementations of the algorithm let the gripper    accessed by wheeled vehicles. There are only few situations
approach the object in a sequence of single steps with inter-     where legs could be advantageous, e.g., climbing stairs or
FSR, Canberra, December 1997                               -4-              Bischoff: Humanoid Mobile Manipulator HERMES

stepping over obstacles. Therefore, we have chosen to equip
the robot base with wheels.
     Ideally, our new robot should be able to move to those
places where humans can go; thus, it should not be larger
than a human. On the other side, it should run independently
from external power supplies and information processing for
a couple of hours. These two requirements are hard to meet
at the same time. An ideal solution could be a robot of cylin-
drical shape. However, this leads to an important reduction
of the loading capacity for batteries, computers and goods,
and it requires specially designed equipment. Therefore, we
voted for a quadratic base with a width of 60-70 cm. This
size exceeds the width of an average human by 10-20 cm, but
still guarantees the passage of doors and navigation in clut-
tered environments humans mostly work and live in.
     Omnidirectional mobility is the most important require-
ment of the driving mechanism. This feature enables the
robot to negotiate narrow passages and to maneuver precisely
near objects that are to be manipulated.

3.2 Manipulation System
To manipulate arbitrarily positioned and oriented objects the      Figure 4: Illustration of the enlarged work space gained by two
manipulation system needs at least six degrees of freedom.         arms (6 DOF each) with two finger grippers, both attached to a
Since the optimal number of degrees of freedom and the             bendable body; the hip joint and the camera platform accommodate
                                                                   for favorable head-hand configurations
kinematic configuration as well as the type of end effector
will be determined in the course of our research the manipu-
                                                                       Whereas humans have limited capabilities to manipulate
lator system should be assembled with standardized modules
                                                                   behind their back our robot should not be restricted in any
that can be combined to form various kinematic chains.
                                                                   way because of his different hip, shoulder and elbow joints as
    According to the anthropomorphic model two arms
                                                                   well as the agile sensor platform. Thus, the robot is enabled
should be used for the manipulation system in order to man-
                                                                   to fulfil visually guided manipulation tasks behind its back,
age also sophisticated tasks, e.g., opening and passing doors
                                                                   e.g., to place objects onto the loading space during periods of
that shut automatically. In addition, many aspects of cooper-
                                                                   extended locomotion.
ating multi agent systems can be studied, e.g., picking up
objects that are too heavy for a single arm ([Lueth et al.
1995], [Khatib et al. 1995]).                                      3.3 Sensor System
    Both picking up objects that lie on the ground and ma-         Sensors are used to give feedback of both interior and exte-
nipulating objects that lie on tables or that are handed over      rior states of the robot system and to supervise the current
by humans are of general interest. In principle, humans can        interaction of its actuators with the environment. Similar to
pick up objects from the ground in two ways: They either           humans the robot’s sensor system can be subdivided into
bend their torso or sink to their knees. Sinking to one’s knees    exteroceptors and proprioceptors. Exteroceptors are excited
can be modeled by a linear drive, but that does not permit the     or activated by stimuli from outside the organism (e.g. eye,
supervision of the grasping process from a sensor head             ears, etc.) whereas proprioceptors relate to stimuli arising
placed above the shoulders. Therefore, we opted for a bend-        within the body (e.g. of tendons, muscles and joints).
able body that brings both arms and sensor head in a favor-
able grasping position (Figure 4).                                 Exteroceptors
    Another advantage over a linear drive is the possibility to    Key to the perception of the robot’s environment is a power-
manipulate objects that are far away from table borders, e.g.,     ful sensor system with the ability to change the focus of at-
in the center of a table. A “hip” joint that allows the robot to   tention actively to desired directions. The design of an an-
bend over its torso at a height of 60-80 cm significantly in-      thropomorphic robot requires a sensor head on a neck-like
creases the robot’s work space depending on the actual arm         joint system placed on the shoulders. At the moment two
lengths and configurations of their degrees of freedom.            degrees of freedom for the neck seem to be sufficient. This is
FSR, Canberra, December 1997                             -5-              Bischoff: Humanoid Mobile Manipulator HERMES

due to the fact that humans use their third degree of freedom    The lowest hierarchical level is formed by micro controllers
just in special cases, e.g., to read letters written sideways    that control actuators and pre-process sensor data. Ideally,
more easily.                                                     these micro controllers should be connected via a standard-
    The neck joint has to be placed and configured in such a     ized bus because the communication with higher level infor-
way that the robot can directly observe the ground in front of   mation processing becomes more effective.
itself. This helps to implement docking behaviors that visu-
ally adjust and optimize position and orientation of the ro-     3.5 Man-Machine Interface
bot’s body with respect to the docking station. The sensor       The design of a man-machine interface has to be divided in
platform needs to be rotated by ± 180 ° to allow the robot to    an interface for the developer of the robot software and an
look behind (e.g. to manipulate behind the back or to drive      interface for the actual user. Basically, the developer needs
backwards guided by vision).                                     multiple display options to be able to supervise and analyze
    The sense of touch enables the robot to gain important       the behavior of the robot whereas the actual user needs a
information for the manipulation of complex objects when         comfortable interface specially adapted to his proper service
visual information is missing, insufficient or could easily be   scenario. An efficient basis for a human-friendly interface is
misinterpreted. In a first step the robot should be equipped     a behavior-based system architecture because it allows com-
with a sense of touch that allows detecting the exact location   munication on a human-like level of abstraction [Graefe,
of collision with respect to the outer limits of the mobile      Bischoff 1997].
base. This enables the robot to learn from incorrect behavior,       An important aspect of both the developer’s interface and
that it exhibited, e.g., during navigation tasks. The grippers   the safety concept is a manual robot control device (e.g., a
could use touch sensors, too, but we do not plan to integrate    joystick) that lets the developer take over control from a safe
them during this first project phase.                            distance. This device must have the capability to control all
                                                                 robot drives directly in order to free the robot from unwanted
Proprioceptors
                                                                 or difficult situations and to bring it manually in predefined
The robot needs various internal sensors that give feedback      configurations.
on its current internal state. Angle encoders on each drive
axis are most important to ensure coordinated motion con-        3.6 Safety Precautions
trol. Torque supervision and overload recognition seem to be
                                                                 The robot should have tactile sensors and emergency swit-
of similar importance and could be realized by motor current
                                                                 ches that disconnect all drives from the on-board power sup-
analysis. Additional sensors should inform about hardware
                                                                 ply in case of danger. If only tactile sensors were used people
failures and the charging state of the battery.
                                                                 would be nevertheless exposed to danger and equipment
                                                                 could be damaged as the kinetic energy of the robot can be-
3.4 Information Processing and Robot Control                     come very high. Therefore, a safety concept must be devel-
Complex robotic systems depend on fast and reliable sensor       oped and realized that early recognizes and reliably avoids
data processing as a basis for intelligent robot control. A      dangerous situations. Such a concept has to be different from
decentralized and hierarchically structured multi-processor      classical safety concepts for industrial robot settings where
system seems to be the best solution for such a demanding        the work spaces of operator and robot are strictly separated.
task. Depending on the required computational power, data        In sharp contrast, most service scenarios depend on a close
links, and available peripherical devices different types of     interaction of operator and robot.
processors suggest themselves for each hierarchical level.           Persons and objects sharing the work space of the robot’s
    On the highest hierarchical level an operator should be      manipulators are exposed to a special danger. Here, safety
able to enter tasks and supervise the whole system. A PC is      can be enhanced by placing tactile sensors on the arm’s sur-
first choice to implement a human-friendly man-machine           faces or by continuously predicting and verifying force and
interface because of the numerous peripherical input and         torque on all joints.
output devices available. On the next hierarchical level oper-
                                                                 3.7 Power Supply
ator commands are transferred to sensor-based actions of the
robot, i.e., motion commands for the mobile base, manipula-      Decisive factors for the employment of service robots are a
tor system or sensor system. A homogeneous network of            long autonomous working time and the decoupling of the
multiple digital signal processors seem to be adequate for       information processing system from momentary voltage
this task because their computational power can be easily        drops caused by abrupt charge changes. To efficiently de-
upgraded by adding more computational nodes as the com-          velop programs a seamless switching between battery mode
plexity of the task or number of degrees of freedom increase.    and external power mode is necessary. Separated emergency
FSR, Canberra, December 1997                               -6-              Bischoff: Humanoid Mobile Manipulator HERMES

switches for motors and information processing system              position. It is easy to modify the
should allow fast and automated start up procedures after          configuration of the undercar-
emergency stops.                                                   riage so that the active wheels
                                                                   are on the left and right side of
4 Realization of HERMES                                            the robot in order to test, e.g.,
                                                                   differential drive concepts.
The humanoid robot HERMES can be described as a multi
robot system with 18 degrees of freedom. These degrees of          4.2 Manipulation System
freedom belong to an omnidirectional base (3 DOF), a ma-           The manipulation system is
nipulation system (13 DOF), and a pan-tilt platform (2 DOF)        mounted on top of the mobile
as sensor head. The manipulation system itself consists of         platform (Figure 6). The arms
two arms with 6 DOF and two-finger gripper each, attached          consist of a structure of double-
to a bendable body (1 DOF) (Figure 1).                             cube shaped turning modules
    Central building blocks of the robot are compact drive         that are connected through con-
modules that incorporate in double cubes powerful motor-           ical and cylindrical linking ele-
gear combinations, the necessary power electronics, various        ments, respectively. Two mod-
sensors (angle encoder, current converter, temperature sen-        ules with an edge length of
sor), a micro controller for motion control and state supervi-     90 mm form the shoulder joint
sion and an intelligent bus interface (CAN) [amtec 1997].          that lacks one degree of free-
With these modules and various mechanical links and adapt-         dom compared to humans. This
ers many different kinematic structures can be built. The          missing degree of freedom can
electrical links for power and communication lines are real-       be partly replaced by a rotation
ized by uniform cables and connectors along the kinematic          of the mobile platform, or can Figure 6: manipulator with
chain of the robot structure. Communication is provided via        be added later if it seems neces- six DOF and two finger
the proven CAN bus. The main characteristics of this bus           sary. The next two modules gripper; length: 94 cm
system are high speed data transfer rates (1 Mbit/s), high         with an edge length of 70 mm
insensitivity against noise, recognition and correction of         form elbow and forearm. A wrist module with two degrees of
transfer errors, multi-master ability and a flexible bus topol-    freedom and a gripper module complete the arm. The shoul-
ogy.                                                               der modules do not extend beyond the width of the undercar-
                                                                   riage.
4.1 Omnidirectional Base                                               Each arm has a mass of 14.2 kg and a payload of 2.0 kg
HERMES is built on a quadratic base of 60 cm x 60 cm, with         (on a fully stretched out arm). By activating the modules’
an additional 5 cm bumper on each side (Figure 5). The             brakes it is possible to exert much higher forces on objects, if
driving mechanism consists of four wheels that are placed in       only the platform’s degrees of freedom are used (e.g., to open
the middle of each side. Two of the four wheels are powered        doors).
                                          and steered, the oth-        The actual proportions of torso and arms yield a grasping
                                          ers are passive cas-     range of 120 cm (!) in front of the robot. The arms can also
                                          ter wheels. Two mo-      reach the rear part of the loading space. The hip module of
                                          tors with a power of     the bendable body is situated at a height of 70 cm. This al-
                                          500 Watts each are       lows for bending the body over tables in order to reach out
                                          sufficient to acceler-   for objects lying, e.g., in the center of the table. Even if body
                                          ate the whole sys-       and arms are fully stretched to the front the vehicle is still
                                          tem at a rate of         balanced because of the heavy and low-lying batteries which
                                          1 m/s2 up to a speed     yield a low center of gravity.
                                          of 2 m/s. The omni-
                                          directional driving
                                                                   4.3 Sensor System
                                          mechanism enables        The camera platform consists of the same wrist module that
                                          the robot to turn in     is used for the arms. The pan axis with a maximum speed of
                                          place and to move        180 °/s is placed on the shoulder and resembles the rotary
Figure 5: HERMES’ omnidirectional
mobile base with active (big) and passive in any direction         degree of freedom of the human neck. The tilt axis (90 °/s)
(small) wheels, bumpers and batteries     from the current         compensates body movements during manipulation and is
FSR, Canberra, December 1997                                -7-                Bischoff: Humanoid Mobile Manipulator HERMES

used to adjust the field of view
during navigation, e.g., to vi-
sually guide docking maneu-
vers.
    Two monochrome video ca-
meras are used for gaining vi-
sual information. It is planned
to add two additional degrees
of freedom for vergence con-
trol and to actively control fo-
cal length and focus. Alterna-
tively, cameras with different
focal lengths could be used
simultaneously to fulfil the
different requirements of the
vision system during naviga-
tion and manipulation tasks.
Color cameras help to separate
objects from the background.
    The proprioceptor system is
mainly integrated in the mod-
                                     Figure 7: Modular and adaptable hardware architecture for information processing and robot control
ules: angle encoders, current
converters and temperature
sensors. Further external and internal sensor equipment will          4.5 Man-Machine Interface
be connected via CAN bus or digital and analogue I/O that
                                                                      The developer’s and operator’s interfaces are realized under
are integrated in some of the modules (e.g., to measure bat-
                                                                      Windows NT 4.0. Tasks can be transferred to the robot via a
tery charge or to realize a sense of touch).
                                                                      wireless LAN. A manual robot control device acts as CAN
                                                                      bus master and is able to control all branched modules and
4.4 Information Processing and Robot Control
                                                                      subsystems using a joystick.
Figure 7 shows the hierarchical multi processor system: The
lowest level is built by the drive modules including their            4.6 Safety Precautions
integrated controllers, sensors and actuators. Each individual
                                                                      To reduce the damaging effect of collisions the mobile plat-
module controls its motion and supervises its state according
                                                                      form is surrounded by bumpers with integrated touch sensors
to commanded parameters.
                                                                      allowing the detection of the point of collision. In the future,
    The main load of information processing is based on a
                                                                      errors of the robot leading to collisions should be analyzed
network of TMS 320C40 (“C40”) digital signal processors
                                                                      and different avoidance strategies should be developed.
forming the second hierarchical level. This is where situation
                                                                          If the robot, especially the manipulators, should run out
recognition, behavior selection, sensor data processing (in-
                                                                      of control, an emergency switch at the rear of the undercar-
cluding image processing) and motion control on a higher
                                                                      riage or at the manual robot control device can be pressed. A
level of abstraction takes place. Here, groups of modules as
                                                                      2-D laser scanner will be integrated in the near future to
functional units (e.g. mobile platform) instead of single mod-
                                                                      allow a robust obstacle detection.
ules are addressed.
    At the moment two C40-based frame grabbers and one
                                                                      4.7 Power Supply
computational node are used for image processing. One node
realizes overall control (including knowledge base manage-            Power is supplied by five batteries which are integrated in
ment and host communication), and a C40-based CAN con-                the free space of the mobile base. Four 12 V batteries in the
troller is used for actuator control and processing propriocep-       corners of the platform are connected in series to provide
tor data.                                                             48 V for the propulsion motors and 24 V for all other drives
    A standard PC serves as host for the multi-processor              and the information processing system. A fifth battery (24 V)
system and realizes the man-machine interface.                        is used to backup the processing system in case of voltage
FSR, Canberra, December 1997                                -8-              Bischoff: Humanoid Mobile Manipulator HERMES

drops. In case of emergency all modules are disconnected            Dario, P.; Guglielmelli, E.; Laschi, C.; Guadagnini, C.;
from the power supply. However, the backup battery still            Pasquarelli, G.; Morana, G. (1995). MOVAID: a new Euro-
supplies the processing system allowing autonomous start up         pean joint project in the field of Rehabilitation Robotics.
procedures for the drive modules and the sensor system.             http://www.alfea.it/movaid/Public_Domain_Area/Papers/Pape
Supply voltages for cameras and other sensors are provided          r1.html, Arts Lab- Scuola Superiore Sant'Anna, Italy.
by DC/DC converters (5 V, 12 V). It is possible to switch           Daxwanger, W.; Ettelt, E.; Fischer, C.; Freyberger, F.;
seamlessly between autonomous (battery) mode and external           Hanebeck, U.; Schmidt, G. (1996). ROMAN: Ein mobiler
power mode. The capacity of the batteries is sufficient for         Serviceroboter als persönlicher Assistent in belebten Innenräu-
several hours of autonomous operation (depending on the             men. In: Schmidt, G.; Freyberger, F. (Hrsg.): 12. Fachgespräch
degree of utilization).                                             AMS 1996, Springer Verlag, Berlin, pp 314-333.
                                                                    Graefe, V. (1992). Vision for Autonomous Mobile Robots.
5 Conclusions and Outlook                                           IEEE Workshop on Advanced Motion Control. Nagoya, pp 57-
                                                                    64.
Based on previously gained experience with work on mobile           Graefe, V.; Ta, Q. (1995). An Approach to Self-Learning
robots and manipulators the humanoid robot HERMES has               Manipulator Control Based on Vision. IMEKO Int. Symp. on
been designed and realized. The general concept of modular-         Measurement and Control in Robotics. Smolenice, pp 409-414.
ity both on the structural and the information processing           Graefe, V.; Bischoff, R. (1997). A Human Interface for an
level assures that HERMES can be used as a flexible and             Intelligent Mobile Robot. To appear: 6th IEEE Int. Works. on
easily extendable experimental platform. By providing two           Robot and Human Communication. Sendai, Japan, Sept. 1997.
arms, a bendable body, and an omnidirectional mobile plat-          Honda (1997). Honda Introduces "Human" Robot.
form a large work space is achieved. The possibility to uni-        http://www.honda.co.jp/home/hpr/e_news/robot/index.html
formly control all degrees of freedom through a standardized        Khatib, O.; Yokoi, K.; Chang, K.; Ruspini D.; Holmberg,
interface and the modular drive concept are sensible                R.; Casal A.; Baader A. (1995). Force Strategies for Coopera-
improvements over existing humanoid robots. We strongly             tive Tasks in Multiple Mobile Manipulation Systems. Intern.
believe that HERMES will enable us to work on many inter-           Symposium of Robotics Research. Munich, October 1995.
esting issues in the field of service robots and will allow us to   Konno, A.; Nagashima, K.; Furukawa, R.; Nishiwaki, K.;
contribute valuable solutions to still unsolved problems in the     Noda, T.; Inaba, M.; Inoue, H. (1997). Development of the
near future.                                                        Humanoid Robot Saika. Proc. of IEEE/RSJ Intern. Conference
                                                                    on Intelligent Robots and Systems, IROS ‘97, pp 805-810,
                                                                    September 1997.
References
                                                                    Lueth, T. C.; Nassal, U. M., Rembold, U. (1995). Reliability
amtec (1997). Produktlinie MoRSE, Beschreibung und tech-            and Integrated Capabilities of Locomotion and Manipulation for
nische Spezifikationen, April 1997.                                 Autonomous Robot Assembly. Journal on Robotics and Auton-
Bergener, T.; Bruckhoff, C.; Dahm, P.; Janßen, H.; Joub-            omous Systems, 14 (1995), pp 185-198.
lin, F.; Menzner, R. (1997). Arnold: An Anthropomorphic             Schraft, R.D.; Engeln, W.; Hägele, M. S.; Kelterer, M.;
Autonomous Robot for Human Environments. In: H.-M. Groß             Nicolaisen, P.; Schäffer, C.; Volz, H.; Wolf, A. (1994). Ser-
(Hrsg.): Fortschrittsberichte VDI, Reihe 8, Nr. 663, Workshop       viceroboter – ein Beitrag zur Innovation im Dienstleistungs-
SOAVE’97, Ilmenau, pp 25-34, September 1997.                        wesen. Eine Studie im Auftrag des BMFT durchgeführt von
Bischoff, R.; Graefe, V.; Wershofen, K. P. (1996). Combin-          April 1993 bis Juni 1994. Fraunhofer-Institut für Produktions-
ing Object-Oriented Vision and Behavior-Based Robot Control.        technik und Automatisierung (IPA), Stuttgart.
Proc. of the Int. Conf. on Robotics, Vision and Parallel Process-   Wershofen, K. P. (1996). Zur Navigation sehender mobiler
ing for Industrial Automation. Ipoh, Malaysia, pp 222-227.          Roboter in Wegenetzen von Gebäuden – Ein objektorientierter
Brooks, R. A.; Stein, L. A. (1993). Building Brains for Bod-        verhaltensbasierter Ansatz. Dissertation, Fakultät für Luft- und
ies. A.I. Memo No. 1439, Massachusetts Institute of Technol-        Raumfahrttechnik der Universität der Bundeswehr München.
ogy, Boston, August 1993.                                           Xie, Q.; Graefe, V.; Vollmann, K. (1997). Using a Know-
Brooks, R. A. (1996). From Earwigs to Humans. To appear:            ledge Base in Manipulator Control by Calibration-Free Stereo
Journal of Robotics and Autonomous Systems.                         Vision. To appear: IEEE Intern. Conference On Intelligent
Cameron, J. M.; MacKenzie, D. C.; Ward, K. R.; Arkin, R.            Processing Systems. Beijing, China, Oktober 1997.
C.; Book, W. J. (1993). Reactive Control for Mobile Manipu-         Yamamoto, Y. (1994). Control and Coordination of Locomo-
lation. Proceedings IEEE International Conference on Robotics       tion and Manipulation of a Wheeled Mobile Manipulator. Dis-
and Automation. Atlanta, GA, May 1993, Vol. 3, pp 784-791.          sertation, University of Pennsylvania, August 1994.
You can also read