Vision Prize 2004 Winner Teemu Sandelin
IRIS – Flexible eye-in-hand servo gripper and tool

Introduction

The inevitable and accelerating trend in consumer electronics, as well as in many other fields of technology, is miniaturization and customization. Automatic assembly systems manufacturing these miniature products will also undergo a scale change and a corresponding performance enhancement in order to adapt to the future’s assembly challenges. Simple downscaling of the current production systems is not sufficient; new technologies in miniature products call for higher accuracy and flexibility. ‘Towards Mini and Micro Assembly Factories (TOMI)’ was a three-year collaborative research and development project at the Tampere University of Technology’s Institute of Production Engineering. TOMI-project aimed to find new solutions for tomorrow’s downsized assembly factories producing tailored miniature products.

Success of part handling contributes significantly to the yield of an assembly line. Machine vision and other sensor based assembly systems pave the way to more agile and flexible part handling and production lines. A high performance and high accuracy Cartesian robot was developed in the TOMI-project. The TOMI Minirobot utilized direct drive linear motors and an eye-in-hand microscope. This new structure is more of an ‘eye-in-palm’ system, and it’s vision based operating principle required completely new control software that integrated the motion and vision control systems.

The need to transfer the expertise developed with the mini-robot environment to more practical form, and utilize existing hardware and minimize costs generated the concept of the IRIS-system, which uses current commercial robots and closely integrates sensor-based control to a versatile eye-in-palm gripper. The concept builds on the centrally located machine vision camera, modular mechanics, ease of use and utilization of the latest machine vision and control technology for minimized size and footprint. The thesis work tried to tackle the task of implementing a ‘plug-and-play’ vision system in the context of mini scale assembly. By creating a part of the context, the gripper mechanics, concurrently with the vision system we are hopefully able to add some easily utilized versatility to the system.

System structure

The IRIS system is comprised of gripper system mechanics, two servo controlled tool axes and a tool (finger) change system, one or two digital cameras, task level control and user interface. Physically the system is divided in three parts: end effector, control unit and PC controller. The end effector is connected to the tool flange of any commercial robot and is the only element that physically interacts with the application environment. The control unit contains a relatively powerful motion controller and amplifiers. The PC is running control software that contains the task descriptions.

The mechanical design produced a small footprint package, which incorporates an eye-in-palm camera, changeable modular actuators, changeable robot interface flange and automatically changeable finger attached to two independent linear servo axes.

Machine vision of IRIS

The IRIS –system’s operating environment sets high demands for the machine vision system. The usability of the whole system is highly dependant of the performance of the machine vision system since most of the desired operations, especially the intelligent ones, are defined by the data gained from the camera view. Non-standard working distance, varying illumination, high accuracy and fast image acquisition and processing were among the challenges for the vision system design. The system now encompasses a compact size digital camera with extended head, mega pixel resolution and a Firewire bus. Considering the application, the most desirable option for camera optics would have been a telecentric lens with a long depth of field in micro lens size scale! The deficiencies of an ordinary micro lens to the system can however be supplemented with alternative task and control strategies. The mechanics and software support also two cameras, and synchronizing two cameras on the same Firewire bus.

The vision sensing can be utilized in many ways during part handling. IRIS tries to combine the traditional search and size measurement tools together with visual inspection tools. The camera can be accurately moved with the robot arm and a plurality of pointers and lighting tools can be attached. There is a possibility to use more specialized measurement methods, such as stereo vision measurement of objects with one or two cameras, localizing the handled objects in 2D or 3D by utilizing the measurement data or by recognizing the parametrically taught edge shape of the object, checking the object and grasp visually, and checking visually that there is enough room for the gripper fingers around the part to be grasped.

For the ease of comprehension the user interface is divided to several windows mostly according to the physical system components, which also partly define the software structure. The work cycles in the programming window are composed of commands and different objects defined earlier in other views, example seen in figure 2.

Each of the objects (robot motion, gripper axis motion, and vision) are configured separately, and then used together with commands to compose work cycle programs. There are commands available even for very low-level control, but most of the programming is meant to be done by using simplified task level commands such as: ALIGN (to) PART1, GRASP PART1, ALIGN PART1 (to) PART2 etc. These example program lines first use vision for locating 'Part1', then grasping it using optimal gripping force, and attaching it to 'Part2'. All operations can be done three-dimensionally. The developed work cycle programming language is an important part of the easy-to-use control software. The software utilizes embedded commercial software libraries for image processing and stereovision applications.

Applications and further development

IRIS-system can significantly improve the ability of a robot to work flexibly and autonomically in an uncalibrated environment. The ’eye-in-palm’ -structure enables direct and continuous workspace measurement and monitoring in every situation, and can hence improve the relative accuracy of part handling operations. For an example in assembly applications the system can support a variety of vision-guided operations, which utilize 2D and 3D measurements in the form of recognition, alignment and monitoring functions. Features visible in the working area can be used for accurately positioning the gripper with the objects in the working environment.

Assembly yield is improved as the vision-guided system can compensate robot backlash in the event of impaired absolute accuracy. Flexible part feeding is easier since no accurate part orientation is required. Even feeding different types of parts simultaneously becomes possible since part recognition can be used to distinguish the parts. Vision based monitoring can help complication recovery without increasing defective work in process. The camera can be used in multiple cell locations for several other machine vision tasks due to its attachment to the robot arm. System reconfiguration is supported by servo axis feature and modular changeable actuators and fingers. The modular design of the system enable attachment to any commercial manipulator even with unsufficient absolute accuracy and any manufacturer specific controller.

Further development focuses on broadening the ‘eye-in-palm’ monitoring for several other tools and applications, including soldering, dispensing and testing. The modular structure of the gripper is currently being expanded to support various tools and their control. Ideally the system would not require an expensive and manufacturer specific cell environment nor robot to operate autonomically. The integrated sensing capabilities together with the modular structure enable automatic adaptation to several different type products at the expense of cycle times and volume.

Considering the continuously changing operational environment and the future of automated assembly business, it is obvious that a need for new agile and low-cost solutions is arising. In the pursuit to decrease still labor-intensive manufacturing and the plant hemorrhage to Far East, it is essential to create systems with similar characteristics as IRIS. Assembly lines with high capital costs and limited flexibility prevent automatic assembly of low and medium level product volumes. New open solutions are required which provide high percentage usage of generic cost intensive components and adapt to small batch sizes and product variations at the expense of unnecessarily high cycle times.

Contact Information: Assembly Automation Laboratory Institute of Production Engineering Tampere University of Technology gsm: +358-(0)400-40 95 69 e-mail: teemu.sandelin@tut.fi www: http://pe.tut.fi/aal