04/11/2009
- Vision system
- Control is key to system flexibility
- Powerful simulation tool
- Fulfilling real demands
- Project partners
ARFLEX
Robots reach new levels of accuracy and adaptability
The precision of industrial robots is influenced by various mechanical effects. Technology developed in the ARFLEX project improves positioning accuracy by 10X compared with previous generations, while providing ample application flexibility to meet the needs of large and small users.
The ability of robotic tools to handle tasks requiring very tight tolerances is limited by mechanical effects such as flexing of the arms, imprecision in the coupling joints, gearbox backlash, general non-linearity and friction. In ARFLEX, a seven-member consortium led by Italian SME EICAS Automazione has succeeded in achieving a 0.1 mm absolute positioning accuracy of the tool centre point (TCP) – a tenfold improvement over earlier systems.
At the same time, it has provided flexibility and ease of use by application of advanced technologies in control theory, embedded systems, sensor devices and vision systems. Added advantages are that the innovative solution is inexpensive and able to operate as a plug-and-play resource in an industrial environment.
Cost-effective approach
In order to correct the positional error of a robot, a suitable sensor system has to be located beyond the active tool, or ‘end-effector’, in order to measure the gap between it and the appropriate reference position. The ARFLEX concept entails direct measurement of both the position and orientation (collectively described as the ‘pose’) of the TCP by means of multiple cameras fixed around the robot, with real-time control to ensure constant accuracy.
The system includes a specially designed contact-less sensor able to measure the pose of a mobile object. Sophisticated self-calibration procedures make this fully autonomous and readily adaptable to different working conditions. As a result, it is possible to use low-cost cameras, without the need for expensive calibration procedures.
Powerful data fusion techniques permit extraction of the necessary information from the multi-camera system, within the timeframe constraint of a 100 Hz sampling frequency. Multi-input, multi-output control loops have also been developed to take account of the non-linearities of the robot and the main causes of error due to elasticity, backlash and hysteresis.
Vision system
Multiple fixed cameras are positioned to view passive infrared-receptive markers distributed on the robot end-effector. A vision system equipped with infrared filters captures only those marker features that are relevant for pose reconstruction. Analysis and testing by the Jozef Stefan Institute verified the overall feasibility and defined the essential component requirements.
The ARFLEX system works in two operating modes:
- start-up calibration mode – using advanced algorithms as the basis for fully automated self-calibration in relation to the robot and the working environment, without requiring any accurate prior knowledge of the camera installation;
- normal mode – where the cameras measure the TCP position and attitude during routine operation, so that the vision control loop can compensate for any error and assure the required accuracy.
To optimise performance under all working conditions, the self-calibration procedures are designed to recover errors due to non-ideal characteristics of the vision sensor and optical lens. Furthermore, FDIR (Fault Detection, Isolation and Recovery) algorithms implemented in the core system enable it to complete a task, even in the event of smart camera failure.
Control is key to system flexibility
Multi-hierarchical control architecture, implemented on a COMAU C4G open controller, provides a good combination of accuracy and flexibility. Importantly, it allows existing robots and their associated control systems to remain essentially unchanged.
The robot’s own controller forms the lowest architectural level, providing commands to the motors for the execution of movements to perform specific tasks. All that is necessary is to make it open to receive and implement reference trajectory corrections sent from the higher levels of the ARFLEX controls, where the vision control loop receives the flow of pose measurements and determines the adjustments to compensate for positioning error.
This structure permits the system configuration to be both modular and distributed, with sensors that can be positioned or changed depending on the application requirements. It also allows further control functions to be incorporated at a higher level. One example is a force compliance control loop developed by research partner IPK, using a software-controlled force sensor to measure external forces acting on the TCP, and to modify the trajectories accordingly. The robot can thus follow profile changes on a surface, or react to forces applied by an operator during a task involving human-machine interaction.
In the real time environment, software modules run on a multi-core central ARFLEX processor under the Linux RTAI operating system. Multithreading programming capabilities provide for parallel execution of the code, and enables the computation time to be reduced.
Communication protocols have been developed for specific channels dedicated to data exchange with the smart cameras, for vision system management and configuration, and for programmed operating task execution via the C4G open controller.
The basic software manages all issues related to communications and interfaces, and forms the bridge between the external devices (robot, open controller and smart cameras) and the internal memories. It is responsible for the real time execution and external data interchanges.
Powerful simulation tool
A crucial part of the research was the construction of a powerful simulation environment, founded on the EICASLAB platform for automatic control design, forecasting and testing.
Algorithms designed and validated in such an environment are compatible with the real-time requirements and are naturally suited for integration into the hardware platform without changing the code. This ability to shift rapidly from prototype to field application makes it a simple matter to transfer control functions to the actual hardware and software components.
The ARFLEX system also accommodates the storage of a large volume of data (system variables, states of robot and control system, host commands, measurements, etc.) that can be used to perform off-line analysis. This feature allows replication of experimental field trials in the simulation environment, where control algorithms can be evaluated and verified in considerable detail.
Fulfilling real demands
By the end of the three-year funded period, three robot demonstrators had been constructed to highlight the project’s outcomes:
- the TCP control system built by EICAS and COMAU illustrates the accurate positioning of a robot end-effector to within 0.1 mm in a 1x1x1 m working space under real industrial conditions;
- a force control and flexibility demonstrator from IPK and COMAU confirms the effectiveness of the force compliance control algorithms; and,
- a visual servoing experimental platform assembled at the JSI premises in Slovenia provides preliminary guidelines for the final ARFLEX vision system design and development.
Today, there is a real demand for industrial robots working at higher accuracies than the best available to date. In aerospace applications, riveting and drilling operations require 0.1 mm positioning – as do metal cutting, milling and deburring in other critical sectors. A 0.15 mm maximum error path is also desirable in sensor-guided robotic laser welding.
For SMEs to adopt robotic production, flexibility and adaptability are overriding considerations. Such businesses tend to manufacture in small batches, for which robots must be readily pre-programmable and easily switched between differing procedures – which is where facilities such as force and compliance controls can be especially advantageous.
The ARFLEX concept meets all of these needs, opening the door to wider deployment of industrial robots to the benefit of both large and small European enterprises.
Project partners
| Actuation and Control Technologies Srl (ACTUA) | Italy |
| COMAU Robotics SpA | Italy |
| EICAS Automazione SpA | Italy |
| Fraunhofer Institut Produktionsanlagen und Konstruktionstechnik (IPK) | Germany |
| Jozef Stefan Institute | Slovenia |
| University of Antwerp | Belgium |
| ZHW Institute of Mechatronic Systems | Switzerland |
Key data
Project type: Specific Targeted Research Project
Project title: Adaptive robots for flexible manufacturing systems(ARFLEX)
Programme: Sixth Framework Programme, Priority 3 – Nanotechnologies and nanosciences, knowledge based multifunctional materials, new production processes and devices (NMP)
Total cost: €4 254 790 – EC contribution: €2 419 000
Project duration: September 2005-August 2008 (36 months)
Coordinator: Dr Gabriella Caporaletti – EICAS Automazione SpA, Italy
More information: http://www.arflexproject.eu
