AS-116.140 Post Graduate Seminar Autumn 1999
Use of simulation in mechatronic machine design
Mechatronic machine design is a very complicated task involving many specialized designers
from different engineering fields. The conventional design methods used in each of these fields
differ slightly from each other. It is difficult to understant the effects of the changes made in the
machine design model to other engineering fields.
Mechatronic machine can be defined as an integration or combination of mechanics, electronics and
software. The power of mechatronics is that this compination can realise more features than any of
the technologies alone. Mechanics can provide physical structure and movement, which electronics
and software can not. Electronics and software can realize logical behaviour in a cost effective way,
which mechanics cannot. Electronics and software are responsiple of the “intelligent” control of the
mechanics. Therefore, behaviour modeling and the integration of behaviour modeling with
mechanical modeling are central to any viable method for mechatronic design.
2 Virtual prototyping
In every design task it is nessessary to build some kind of a model of the system to be build.
Software designers use CASE-tools for graphical software modeling and electronic designers use
logic-diagrams and time-diagrams etc. In mechanic design graphical drawings and models are not
sufficient enough. Some kind of real physical verification of the model is very often needid allready
in early steps of the design. Traditionally it has been used physical models, mock-ups or physical
Designed machine model can be tested against real physical constraints, against forces and
movement limitations and many other environment conditions using physical prototypes.
Prototypes are made not only to investigate the operation of the machine but also the ergonomics,
aesthetics and assembly of the machine.These prototypes can be however difficult to build and time
consuming and changes are difficult to realice. If other parts of the mechatronic machine, for
example software or electronics are not yet in prototype phase, testing the full behaviuor is not
possible with physical prototypes. That is why virtual prototyping is coming more and more
important in mechatronic machine design. Cheap and effective computer platform development has
made virtual prototyping economically possible also in smaller scale design projects.
Virtual prototyping is a design tool which suports the design prosess in the early stages, to minimise
the associated risks before a large investment in continued development is made. The early
introduction of virtual prototypes will give designers and multidisciplinary teams greater freedom to
explore trade-offs and the extremes of variation. In its most basic form a virtual prototype will
provide a three dimensional visualisation that will enable the teams to establish a shared vision of
the design and identify potential conflicts. With the addition of intelligent geometric reasoning,
constraint management and dynamics simulation techniques, the virtual prototype becomes a very
powerfull interactive analysis tool.Definition of virtual prototyping
Virtual prototyping can be defined as a software-based engineering discipline that includes the
modelling of a system, simulating it and post-processing results. The modeling of a system means,
in fact, creation of the set of equations describing the system being studied. The simulation is
numerical solving of the created equations as a function of time. Post-processing the results means
visualizing the behaviour by using traditional diagrams or more illustrative 3D-animation.
(comment: In fact there is interconnection between the last two parts so that graphical models are
needed also in solving the equations in the solver (contact modeling and simulation)).
1. Modeling of the system
In mechanical design especially in machine assembly design realistic and accurate geometric 3D
model is important. The machine assembly prosess can be evaluated if constraints can be imposed
to the geometric model. This is one aspect of virtual prototyping. So we can list three fundamental
consepts that belong under term virtual prototyping:
• Photo realistic visualization
• Constraints imposed on the geometry (assembly)
• Simulation of physical behavior (gravity, collisions and friction etc.)
Definition of Simulation.
Simulation is the process of building and experimenting with a computerized system model such
that a specific purpose of the study is achieved through observing model’s behavior under
assumptions defined by the experimenter.
3 Different types of simulation packages
Simulation and modeling of systems can be done with many types of software programs. It can be
done even with general purpose standard programming languages but it is not an effective way.
There are higher level simulation languages which are specifically designed for simulation
applications.(Ada, ADSL, SIMULA, MIMIC, CSMP etc.). However for an effective virtual
prototyping purposes simulation packages are needed which support interactive mode of simulation
and which have good graphic features for visualisation. (2D 3D, animation, VE, VP)
Continuous time virtual prototyping simulation programs can be divided into two main types.
• General continuous time simulation programs
• Simulation programs for special engineering fields.
The first category includes general simulation programs (Matlab/Simulink). These programs
usually solve nonlinear time dependent systems of differential equations but the equations to be
solved have to be derived outside the simulator code. The advantage of general mathematical
programs is that they are very flexible and all kinds of systems can be modeled. Usually also
limitations of variable values, the time step and error tolerance in integration algorithms can beadjusted. The disadvantages are ineffective integration algorithms and difficulty in constructing
larger models. There is big error making possibility and also changes are difficult to make.
The second category includes simulation programs that have been designed for a certain area of
engineering, for example industry robot simulators and car design simulators. (Adams, Dats,
impack, Ideas, Envision). These programs have preprocessors which are used to generate the code
for the solver. The preprosessor usually has libraries for basic components of the systems to be
simulated. The simulation model is composed by connecting these components and defining their
dependencies. The advantages are that user don’t have to derive the equations of motion. The
kinematics and often also dynamics is available. These programs are very interactive menu-driven
so that changes to the model and diffrerent parameters are easy to make.
A modern mobile machine includes a number of engineering areas, such as mechanics, hydraulics
and control electronics. All these have their own simulation programs, but a combined simulation of
a complete machine system is still very rare.
In mechatronic machine design mechanical dynamics simulation is done with one simulation
program and structure flexibility is modelled with an other program and if thermal analysis of the
motor for example is needed that is modelled with third program. For mobile machine also ground
contact simulation possibility is important.
Another way to divide packages is according to their graphical capapilities.A Good Mechatronic
Prototyping Software has to import CAD models of different types (Mechanical Desktop,
Intergraph Solid Edge, SolidWorks, PRO/Engineer) and to show 3D virtual environments with
• Only numerical data input and output as graphical diagrams and plots.
• 2D graphics with animation
• 3D graphics with animation
The interactive mode of simulation means that the simulation process can be interrupted for the
purpose of asking, or reporting to, the user. The user can make fast modifications to the model and
simulate again.4 Connections between simulation software
Program that can do all the different simulations needed in virtual prototyping process does not
exist. That is way connections between different software packets are needed. Packets have to have
transformation modules that can convert between different representations and file formats used in
different simulation programs and CAD-programs. Possibility to connect user defined function
libraries to simulation program is very important. See Figure 1
Testing of the actual mechatronic machine modules with help of the simulator is also possible if the
simulator supports low level connection to networks so that control can be send to machine
controllers and machine sensor data can be directed to simulation software . See Figure 2.
CAD -program Simulation -
FEA -program simulator
Thermal Result Data
simulation - Analyse -
Figure 1 Connections of different software packets around virtual model.
5 Few concepts
Virtual prototyping program with Virtual Environment has usually interactive physics-based, 3D
graphic models to visualize and evaluate concept designs for complex systems and subsystems. The
CAD-based models represent the actual geometry and motion characteristics (kinematics only)
associated with the real world system.
Virtual Prototype simulation
Virtual Environment can be enhanced by incorporating dynamics into the physics-based models to
create a virtual environment that supports simulation based design. In this environment, users are
able to design, build, test, operate, and support multiple product and system scenarios.
Virtual prototyping program which supports virtual reality provides a environment for designing
and programming complex design, engineering, and training operations by allowing the user tointeract more directly with the simulation environment. Users can fly and walk through the
simulated environment using virtual reality and telepresence techniques to operate devices, alter the
environment by relocating components, or otherwise interact with the virtual world. This needs
special connections to different kinds of VR-interface devices.
Figure 2 Simulator low level connection to physical machine prototypejoint values: Tool coordinates:
Tool coordinates: joint values:
Figure 3 Direct and inverse kinematics
6 Dynamic machine simulation
6.1 Simulation of machine kinematics
The simplest motion simulation used for example in robot off-line programming, is only graphical
simulations based solely on a kinematic analysis of the machine models. In this approach, motions
are planned to satisfy positioning requirements in device configuration space and/or workcell
Cartesian space. The simulation engine operates on the assumption that each device can precisely
follow (or track) the motion planning commands in time, without regards to the forces and torques
that cause the motion. The resulting behavior of the Workcell is thus idealized, in the sense that all
devices move along their planned trajectories for all time.
6.2 Simulation of machine dynamics
Although a kinematic simulation is perfectly adequate for many problems arising in robotics, it is
generally not appropriate for applications such as prototype arm analysis with different work loads
and controller design. In these cases, the dynamics of a particular Device must be analyzed to
ascertain the feasibility or effectiveness of an engineering design. A sound dynamics simulation can
provide the engineer with valuable insight into the problem at hand, as well as experimentally
verify the operation of a prototype in the various test loops.
Newton - Euler dynamic equations are commonly used.in dynamic analysis (see Equation 1).
Hamilton principle and Lagrange dynamic equations are an efficient way to solve device dynomic
f = m a
n n n
R C R R C R
τ = I α + ω× I ω
n n n n n
Equation 1 Newton - Euler formulation
Two basic solutions to the dynamics problem:
1. Device Dynamic Analysis
2. Device Dynamic Simulation.
Device Dynamic Analysis is suitable for prototype arm design and motion design analysis. This
capability, which solves the so-called inverse dynamics problem of robotics, computes the Joint
torques required to produce a planned motion trajectory.
End effector End effector
force trajectiory position trajec to ry
Inverse Fo rw a rd Fo rw a rd
Ja co b ia n dy nam ic s k inem atic s
J oint Jo in t Po s itio n
to r que s positions
τ θ ω α
n n n n
Ja co b ia n
dy nam ic s k inem atic s
tim e tim e
Figure 4 Dynamics and inverse dynamics
Device Dynamic Simulation is intended for applications such as controller design and trajectory
tracking simulations. This study computes and displays the motion trajectory of a Device, given the
control torques exerted by its actuators (the Motion Controller), and the physical properties of the
Device itself (the Physical Model). The simulation encompasses the so-called forward dynamics
problem, which describes the dynamic behavior of the Device in terms of the time derivative of the
arm configuration. It is important to note that dynamic simulation is quite different from dynamic
analysis, in the sense that simulations will graphically show the trajectory of the Device in response
to its control law, i.e., with the concomitant overshoot, oscillation, and tracking error.Figure 5 Device with three link parts and two rotational joints
7 Phases in machine model building
To get an idea of what kind of phases there are in building and simulating a dynamic machine
model brief list of operations needed in one quite common simulation software packet
7.1 Part model
• Importing 3D CAD drawings of the machine parts to simulation programs CAD module.
• Adding base coordinates and auxilary coordinates according to Denavith-Harttenberg rules.
• Giving color attributes to parts
• Saving part objects in systems own representation format.
Device consists of link parts and joints connect these parts. For the purpose of dynamic analysis and
simulation, a Device is defined as a collection of rigid bodies, henceforth called Links, joined by
single degree of freedom rotational or translational Joints, i.e., no shear or scale. Each Link may
consist of just one Part, or multiple Parts fixed in relation to one another. The user must specify the
mass parameters of each Link in the Device.
7.2 Device model
• Picking up the base part from part file
• Creating Device with selected part as the base part
• Setting main properties of the Device.
• Attaching new child parts to the parent parts coordinates
• Saving the Device model7.3 Main properties of the Device
• Name of the device
• Degrees of freedom
• Is dynamics used with this device or not
7.4 Special properties of the Device
• Number of IO-signals between Devices
• Number of home positions
• Number of Uframes
• Number of Utools
7.5 Defining the device kinematics
• One joint can have only one degree of freedom (dof).
• Assign the type of dof for the joint. It can be translation Tx, Ty, Tz or rotation Rx,Ry,Rz.
• Simple kinematics means that device can have all the six dofs (Tx, Ty, Tz , Rx,Ry,Rz ) without
any defined joint, so that the device can move to all directions and rotate around its three axes.
Simple kinematics is used for example in AGV- device models. (Autonomous Gaided Vehicle).
7.6 Defining the inverse kinematics type
• Generic type: the program guess the inverse kinematics type from the device structure ()
• Numeric: means that program calculates the solution through iterative calculations.
• Device type: inverse kinematics type of same kind of device is picked from library
• User type: User has to write the algoritm code.
7.7 Defining the device attributes:
• Home position
• Motion limits, like max accelleration , max velocity, max torque
• Display attributes
7.8 Defining the device dynamic attributes:
• Mass, mass center point and inertia tensor of each part
• Coulomb friction and Viscous friction of each joint• Spring constant for each joint
• External forces
• Gains for each joint
7.9 Settings needed before simulation
• Controller settings with heart rate.
• Integrator selection for dynamic equation solver with step size.
• Simulator step size.
• Motion profile selection (standard acceleration/deceleration, corner rounding, constant velocity)
• Display step size8 Simulation of the model
8.1 Simulation program
When we want tosimulate the machine model we make first a program to move the device and its
parts the way we want it to move in the virtual environment. This program is made using specific
language which is called Grafical Simulation language program (GSL) in Envision (Deneb)
When we start simulation we start this program. In every movement command statement in the
program control jumps to the so called motion pipeline. When movement is ready and the target
position reached in motion pipeline loop the display is updated and the control goes to the next GSL
8.2 Motion pipeline.
Figure 6 Motion Pipeline Flow of Control8.3 The Motion Model
The Motion Model is responsible for modelling the motion of Devices under the action of forces. It
is essentially the last computational unit in the Motion Pipeline, see Figure xx.
The Motion Model consists of a series of procedure calls that are executed sequentially during the
simulation of a Device. Each component in the pipeline is dedicated to a particular problem in
Figure 7 Motion Model
During a simulated move, the Theta Generator evaluates the planned trajectory of the Device for
the given eval_time and current motion type. This computation results in the desired Joint DOF
positions, speeds, and accelerations (i.e., planned_thetas). The next step executed in the pipeline
depends on the current mode of Dynamic analysis.
In the case of Device Dynamic Analysis, the Torque Analysis module computes the current_torques
required to achieve the planned trajectory. The interlink forces and moments are also calculated ateach step. Control then passes to the Identity Model, where the current positions, speeds, and
accelerations of the Device Joints are set equal to the planned values, i.e., current_thetas =
planned_thetas. This sequence of events is repeated at a period equal to the Simulation Step Size.
In Device Dynamic Simulation mode, the Motion Controller begins by computing the
current_torques based on the following parameters:
• the control law coded in the Motion Controller,
• the current trajectory of the Joint DOFs, i.e., current_thetas, and
• the Joint DOF trajectory error, i.e., planned_thetas -- current_thetas.
8.4 The Physical Model:
The vector of Joint torques is then input to the Physical Model, where it is immediately filtered by a
saturation function which models how much torque the actuators can actually deliver. The resulting
motor torque values, along with the current state of the Device, are used to determine the time rate
of change of the Device state variables. Finally, the latter quantities are integrated to obtain the new
state of the Device. To complete the simulation procedure, the current Joint positions, speeds, and
accelerations are updated to reflect the new state. The process is repeated at the next Controller
Program provides two built-in forward dynamics modules:
1. Recursive form of Hamilton's equations of motion. O(N)
2. the Newton-Euler formulation. O(N3)
( O( N)=dependency on the number of multiplications and additions)
8.4.1 Integration of equations of motion
The equations of motion of a Device are integrated using standard techniques from numerical
analysis. The default integration algorithm is a second-order Adams-Bashforth method, which is an
efficient algorithm utilizing state information from more than one previous time point. It is accurate
enough for animation purposes but It can be used only with devices having only moderate stiffness.
1. Adams-Bashforth method
4. Runge-Kutta-Fehlberg methods.
In general, the 2 and 3 of integrators are used when extreme accuracy is required. Adams-Moulton
algorithms are used to simulate the dynamics of stiff mechanical systems (Devices with large
friction coefficients or spring constants). Selection of the right integration method is very important
in order to have accurate and stabile solutions.8.5 Motion Control
The default Motion Controller product implements an independent-joint PID control law. In the
algorithm, the applied actuator torques are set to a value proportional to the Joint trajectory error,
with an additional term included to compensate for friction (both viscous and Coulomb). Note that
aside from frictional effects, this control scheme completely ignores the plant dynamics.
In order to get good results in dynamic simulation comparable to physical prototype tests it is not
enough to have good physical model of the machine, but also the motion controller module should
be the same or equal as used in the real physical prototype.9 CASE: Simulation of autonomous mobile robot.
The new outdoor/indoor platform for service robot called WorkPartner alias Hybtor.Hybtor is new
test platform for mobile robotics both outdoor and indoor environments. It uses a hybrid locomotion
system which combines both wheels and legs to produce active propulsion. Because of its four
wheeled legs and articulated body it has very flexible and wide range of mobility.
HYBTOR-platform will have four 3DOF legs, a controllable body joint and a wheel in each leg
which make it capable of locomotioning flexibly on almost all type of terrain. The power system
uses batteries and a small 4 kw engine. The overall weight is about 160 kg and planned payload
about 60 kg. The computer system will be based on CAN, microcontrollers, and PCs.
Perception system (not shown here) includes stereo vision, laser for environment perception and
inertial and force sensors for body motion sensing. The interface with the operator is a highly
interactive multi-media interface which makes it possible for the
operator to communicate on the cognition level with the robot.
9.2 Modeling the robot
Figure 8 One leg of the mobile robotSimulating movements in the virtual environment
9.3 Motor torque analysis
• Analyzing torque needed in joints using different control and motion
trajectories and velocity profiles.
9.4 Robot posture and gait study
9.5 Study of the robot movement on uneven terrain
• Walking and wheel driven movement
9.6 Ground contact modeling
9.7 Study of the mobile robot balancing control
9.8 Position and force control
9.9 Climbing control
9.10 Dynamic walking controlFigure 9 Movement and forces of one leg on different terrain
9.11 Validation of the robot software modules
Motion planer and study of smart trajectory strategies
Inverse kinematics module validation
Motor feedback-controller module validation
Figure 10 Inverse kinematics module test9.12 Validation of the mechanical system
Ethernet Control data Sensor data Camera data
It is possible to test the mechanical system before the high level motion
control software is ready. Tester moves the simulation model using simulator
interface and the wanted joint position commands are send to the real robot
through Ethernet. Feedback data from the robot, the actual positions of the
joints are fed back to an other simulation model so that the actual robot
motion can be compared and analyzed with respect to the motion commands
sent to the robot.
If the simulator and operator are in different rooms than the actual robot,
camera control and monitoring of the robot should be added to the virtual
Figure 11 Hybtor turning using joint in the middle.10 CASE: Simulation based design of a wheelchair mounted robot
Gunnar Bolmsjö et.al /6/ have used simulation tools in designing a wheelchair mounted
robot for disabled people. In this kind of special robot design, which is very different
from standard industry robots, unfamiliar new ideas and design concepts has to be
developed. Robots with higher degree of autonomy and intelligence is a modern trend.
Robotic design has to go in that direction so that ordinary people , users of the future
robots, are able to communicate with them and use them. This trend can be seen from
new robotic application areas coming up. One of these areas includes health care and
assisting disabled and elderly people.
The primarily user in mind of the designers was typically a high spinal injury. The result
of such accidents is that the individual has no controlled finger motion and possibly only
limited upper arm motion enough to control a joystick to drive the wheelchair. It
important to adapt to different requirement related to the manipulator and the user
interface and input device.
10.2 New design concept
Robots and humans have to work together in a common workspace and doing the same
tasks when helping disabled people. They have to work like work partners, because it is
not possible to build one intelligent robot that can do all the necessary things needed for
example in household. Examples of tasks the robot can help are: control of different
equipment through push buttons such as, computers, copy machines, turn on/turn off
lights, functions in the kitchen, etc.
In this kind of robot design different specifications compared to those used in industrial
applications has to be considered. Examples of differences are:
• the payload/weight ratio must be far higher than today's robots giving priority to
movability and quick set-up at new workplaces,
• a larger workspace and a more flexible configuration will be needed compared to
industrial robots of the same size,
• acceleration and velocity performance can in general be much lower compared to
• design criteria must enable production at a low price in high volumes.
The basic needs of a large working range and a low weight of the robot forced to try
non-conventional design concepts. The design group used simulation tools (Deneb/Igrip)
allready in the early phases of design. Simulation tool was used for fast simulation of
ideas coming out from brainstorming at meetings.
One basic characteristic feature is the use of rotating joints having a 45-degree angle
between the input and output axis. During early simulations this was found to give a large
working envelope together with relatively low torque and possibilities to fold themanipulator in a small volume. Some sample studies from simulations in Igrip are shown
in Figure 12
Results from early simulations
• rotating joints having a 45 degree angle between the input and output axis
• a large working envelope
• relatively low torque
• possibilities to fold the manipulator in a small volume
Figure 12 . Some sample studies from simulations in Igrip /6/
Detailed design analyzes through simulations
After the initial simulations which visualized different working scenarios of the robot arm
a detailed simulation study began. The purpose was to validate the possibility to construct
and assemble the robot arm with new kind of skewed joints. After this simulations were
made to minimize the weight and the torque needed to drive the arm.
10.3 First physical prototype
A three axis prototype was built with motors in the base and transmissions through the
arm and harmonic drives in the joints.The main draw-backs that was found:
• Weights too much
• Safety reasons
• The robot arm must be able to drive by hand in case of a break down for safety
10.4 Change of the design direction
totally different direction was chosen with integrated drive systems inside the robot arm.
In conclusion this demand for integrating the following components for each joint and
1.Minimizing the weight of each part using light materials.
2.Integrate the drive systems in the joints.
3.Integrate the motor controllers for each rive unit in each link to minimize the cabling
in the robot arm.
• high performance brushless motors
• small motor control unit for servo control
• intelligent'' distributed control using CAN-bus.
Precise simulations to evaluate different designs.
These demands could not be worked through without further analyzes through
simulations in Igrip. Specifically, precise simulations started to evaluate effects of
different designs with respect to required torque of the drive units for a ``standard'' work
cycle defined as a simulation case in Igrip.
10.5 Simulation of different cases for a suitable place for the robot and
first working prototype
The new design concept led to new limitations in the final construction as well as
demands from the available space on the electrical wheelchair. It was difficult to find a
suitable place for the robot. Therefore, simulation of different cases was a necessity in the
design and construction work.
As the project had to go in a new direction there was a need for a quick analyze of
possible solutions which could work for a first working prototype.
Results of the designThe results from extensive simulations and analysis in Igrip together with design
modeling in PRO/Engineer in an integrated environment was an eight-axis robot arm
which is modular and configurable for different applications.
Through the modular approach the same components can be used to configure a quite
different robot arm for use as a stationary robot for assisting tasks.
10.6 Virtual scenarios
The virtual scenarios were used to validate the functional specifications of the robot.
Simulations were done to show the use of the robot in situations similar to reality. In the
simulations, real environments are modeled which include texture mapping to enable
high degree of realism. So that it would look the same when real tests start in the physical
world, see Figure 13 realistic virtual scenarios provided a possibility to discuss user
requirements and user interfaces with the intended users before the robot exist.
Figure 13 Simulation of a virtual scenario in an office environment /6/.10.7 Validating the mechanical robot arm
At an early stage in the project the control system will have limited functionalities. At
this stage the Low Level Interface is used to send robot joint data from Igrip to the real
time controller which receives the data and drive the robot to the target position.
Through successive software development the different modules can be tested in Igrip
such as kinematics and motions controllers, before implementing them in the real time
To drive the servo motors from simulation program igrip the joint data (angles) are sent
to the real time controller which coordinate the control of the motors through its motor
controller and PID regulators.
10.8 Developing and validating the control software modules
Testing and validation of the performance of the motion related software modules such
as kinematics, motion planner, smart trajectory strategies was used through Igrip .
An Artificial Neural Network was trained to solve the inverse kinematics of the
manipulator to work in a specific work space. It was noted that for this work Igrip was a
fast tool to validate the performance of the inverse kinematics.
10.9 Validating the control system
To validate the controller feed-back data from the robot (motor feed-back) the actual
position of the joint angles are fed back to Igrip where the robot motion can be compared
and analyzed with respect to the motion commands sent to the robot. Test groups opinion
was that the simulation tool helped a lot in debugging the performance of the robot. It
could be used in this way to simultaneously develop both the mechanics and the
controller software./ 6/11 References:
1. M.R.Thompson et.al. Interactive Virtual Prototyping. Keyworth Institute of
Manufacturing. Univ. of Leeds. http://www.scs.leeds.ac.uk/kwb/IC/vp.html
2. A.Rouvinen et.al. Virtual Prototyping of Mobile Machines, State-of-the-art.
International Conference on Engineering design. ICED 97 Tampere, Aug. 1997.
3. Ju-Yeon Jo et.al. Virtual Testing Of Agile Manufacturing Software Using 3D
Graphical Simulation. CAISR technical report TR-96-107, submitted to IEEE
4. B. Chan and Susan Finger. Bringing Virtual And Physical Prototyping Into The
Classroom. International conference on engineering design. ICED 97 Tampere, Aug
5. Dynamics Reference Manual. On-line document by Deneb 1999.
6. Gunnar Bolmsjö et.al. Simulation based design of a wheelchair mounted robot for
disabled people. Proceedings of Deneb User Conference, Troy (Detroit), USA,
October , 1996.