|main site > home > Personal Robotics Seminar, London 1985 >||28 May 2007|
Personal Robotics Seminar 3rd July 1985
Developing a Personal Robot from Concept to Final Product Tim Jones, Technical Director of Universal Machine Intelligence Ltd.
This paper describes UMI's R Theta mobile robot and its subset, the RTX stand alone arm.
Between the "Star Wars" inspired toy robot market and the industrial robot there lies a grey area called "Personal Robotics".
As with any new product category, definitions of what constitutes a personal robot are open to dispute. As a result the term has become a catch-all for all sorts of non-industrial robots whether costing a few hundred dollars or many thousands.
If an analogy is made with computers, then the industrial robot equates with the minicomputer, being a significant investment, the toy robot equates with the home computer, being for entertainment and the Personal Robot with the Personal Computer a la IBM PC.
The point to note here is that personal computers are useful, productivity increasing tools that are cheap enough not to require detailed cost justification.
Likewise, Personal Robots should be fundamentally capable of doing useful work at an affordable price, even if in practice they are limited by the currently available technology.
At UMI we believe in this principle very strongly and our product design and development strategy reflects this.
The Size of the Job
Producing a low cost but useful robot is a difficult task, particularly if it is a mobile unit and trying to produce an "all singing all dancing" domestic robotic slave, using today's technology, is doomed to failure . This is not to say that the robot "maid" will never appear, the issue is when, not if.
It is this vision of the future which tells us the size of the job and it gives us hints about the products we should be producing today.
To be acceptable to the general public we can envisage a robot which ultimately responds to a verbal command such as "Fetch me a beer from the fridge" and this assumes:
1. multi-user continuous speech recognition with natural language understanding resulting in; 2. task interpretation, which presupposes the existence of; 3. a knowledge base of objects with appropriate descriptors and; 4. a World State database describing the relative positions of these objects with; 5. rules concerning the objects and their interrelationships, which with; 6. an Inference Engine and Task Planner produces; 7. a Plan with goals and sub-goals.
This plan requires that the robot:
A. knows where it is and what orientation it is in. ie it requires some form of navigation system; B. knows the status and position of its arm(s), implying closed loop control of its joints; C. has effective mobility that can be adequately monitored by the Planner, ie closed loop control with status reporting; D. has sensors for obstacle detection, object location, discrimination and identification. Probably a combination of Ultrasonic ranging, vision, touch etc; E. has an arm and end effector(s) capable of opening doors, manipulating useful loads (several kilogrammes) and moving accurately in Cartesian coordinates; F. Has some form of feedback to the user for command clarification, probably speech synthesis with rules for phrase and sentence construction; G. Has a user friendly method of obtaining and updating its database.
Clearly to try and produce such a beast from scratch would require a massive investment over a long time period and by no means could commercial success be guaranteed.
So what is the answer?
The Design Approach
Two words underline our design philosophy;
MODULAR and EXPANDABLE.
Rather than trying to tackle all of the problems straight away, we have concentrated on those areas which are essential to a robot system and are least vulnerable to changing technology. These are:
1. The manipulator. 2. Mobility. 3. An Expandable Supervisory Computer.
By making sure that the system has the necessary expansion hooks, the robot can grow as the technology allows .
For the arm to be useful we felt it it should be able to :
1. Lift and manipulate objects up to 2 kilogrammes. 2. Be able to pick up objects from the floor and from desktops and have a reasonable reach. 3. Have modest power consumption with no power required while the arm is stationary. 4. Be compact when not in use and visually attractive . 5. Have a general purpose end-effector capable of picking up large and small cylinders and parallel sided objects. 6. Be modular in itself allowing one, three, four and six axis possible options with interchangeable gripper. 7. Be reasonably rigid and with low backlash. 8. Have closed loop servo control, with manual and numeric position control, velocity profiling and force control, with full status and error reporting. 9. Have a mechanical structure that lends itself to operation in cartesian coordinates.
After considering several arm configurations a SCARA format with extended vertical travel and three axis wrist was settled on.
The main arm sections swing in the horizontal plane and therefore the motors driving them do no work against gravity.
The wrist motors drive through a spiroid bevel gearset to give pitch and roll. Both these motors work in tandem, driving in the same direction to give pitch and in opposite directions to give roll.
Backlash is contolled on the wrist by the use of shims and by the use of two stage belt drives on shoulder, elbow and wrist yaw.
Vertical drive is achieved using a geared motor driving a timing belt which moves the arm carriage along a custom designed aluminium extrusion.
Motion is controlled by the use of five pairs of rollers running on three tracks with eccentric adjusters controlling running clearance. A ribbon cable runs to this carriage taking power and signal wires from the arm controller and this is covered and protected by a cosmetic blind.
The arm operates in a truncated cylindrical envelope of over twenty cubic feet principally in polar coordinates and can reach in front of and behind the body of the mobile robot.
The upper and lower arms have the same length and there is a 2:1 difference in gear ratio at the shoulder and elbow. Due to this, radial motion can be achieved simply by driving the shoulder and elbow motors at the same speed but in opposite direction. Angular rotation is achieved by driving the shoulder alone. It is this characteristic that gives the mobile unit its name- R Theta.
Yaw is driven from a motor in,'the upper arm through an intermediate combined pulley on the elbow spindle. There is a 2:1 belt driven reduction from this pulley to the wrist pulley. This means that the wrist remains in the same orientation to the radial line passing from the shoulder pivot to the wrist pivot, without yaw being driven.
This allows the arm to be controlled simply but effectively.
Control of the seven DC motors of the arm and the two base motors is via two 8031 processors. Feedback from the motor encoders (in quadrature) interrupts the processors and these monitor position and direction of all the axes.
These two intelligent peripherals (I/Ps) reside on a double eurocard and communicate either along an RS232 link to an external computer, if the arm is used as a stand alone device (RTX), or through a TTL link to the system backplane to our Master Supervisory Processor (MSP).
A PID control algorithm is used with velocity profiling to drive the arm in unbounded Manual mode (drive until told to stop) or in absolute or relative Numeric mode (where angles and displacements are specified.) An interpolation mode is also avaiable for continuous path control. Force thresholds can be specified both in position and force control mode. Power to the motors can be switched off in free mode and the status and position of the arm can be accessed at all times.
Four, levels can be defined;
1. Joint level (or motor level) 2. Manipulator level, where position and orientatior of the gripper can be specified using inverse transformations. 3. Object level, where routines can be specified and relocated in the workspace. This requires matrix transformations . 4. Task level. This is where only the task is requested and the task planner plans the robot task sequence.
The first two levels will be available with the RTX standalone arm with software available on an IBM PC. The first three levels will be available with the MSF for both RTX and R Theta and the fourth if an additional task planning processor is added to the system.
Mobility can be achieved in a number of ways, namely:
1. wheels 2. tracks 3. legs.For simplicity we have opted for a tricycle configuration with two driven wheels and a castor.
With closed loop DC servo control this gives good straight lines and arcs of any radius can be specified by differential speed control of the wheels.
Master Supervisory Processor
The central control of R Theta is by the MSP, an 30C88 based card with 8087 coprocessor option, running a Multitasking Forth (Added Dimension Forth by Bill Stoddart).
This board drives an expansion bus and robot serial bus through a seven slot backplane. A combination of up to 8 intelligent peripherals can be accessed or the serial bus and 8 expansion bus peripherals. These expansion bus peripherals can be memory, I/O, or intelligent cards such as vision, sensors etc with their own local processor.
The Forth is reflected through the 16 64kbyte pages of memory allowing memory residing on the expansion card to hold Forth words specific to that card. This allows the Forth language to extend beyond its normally constraining 64k.
The MSP has an RS232 link to an external computer, an infra-red remote control input and a multiplexed joystick port. Memory is battery backed static RAM.
The first expansion card has a calendar clock with wake up facility, MEA 8000 Formant speech synthesiser with audio amplifier, Memory interface for ROM packs etc and power monitoring and control circuitry.
Following this we will be producing other expansion cards including gripper vision and a line following navigation system. Clearly the expansion doesn't stop here. It is down to third party developers as well as ourselves to develop expansion options, software and potential markets. This is what the product is all about.
A significant amount of technology has to go into a personal robot if it is to be a truly useful device. This will not happen overnight, but with a robust and carefully designed modular product that has the necessary expansion capability, progress can be made.
There is a threshold where people can see the potential of a personal robot and can get excited by the very real prospect of it doing something useful. When this is reached, the market will explode.
At UMI we intend to be there when it happens.