For several practical control engineering applications it is desirable that multiple systems can operate independently as well as in cooperation with each other. Especially when the transition between individual and cooperative behavior and vice versa can be carried out easily, this results in ??exible and scalable systems. A subclass is formed by systems that are physically separated during individual operation, and very tightly coupled during cooperative operation. One particular application of multiple systems that can operate independently as well as in concert with each other is the cooperative transportation of a large object by multiple Automated Guided Vehicles (AGVs). AGVs are used in industry to transport all kinds of goods, ranging from small trays of compact and video discs to pallets and 40-tonne coils of steel. Current applications typically comprise a ??eet of AGVs, and the vehicles transport products on an individual basis. Recently there has been an increasing demand to transport very large objects such as sewer pipes, rotor blades of wind turbines and pieces of scenery for theaters, which may reach lengths of over thirty meters. A realistic option is to let several AGVs operate together to handle these types of loads. This Ph.D. thesis describes the development, implementation, and testing of distributed control algorithms for transporting a load by two or more Automated Guided Vehicles in industrial environments. We focused on the situations where the load is connected to the AGVs by means of (semi-)rigid interconnections. Attention was restricted to control on the velocity level, which we regard as an intermediate step for achieving fully automatic operation. In our setup the motion setpoint is provided by an external host. The load is assumed to be already present on the vehicles. Docking and grasping procedures are not considered. The project is a collaboration between the company FROG Navigation Systems (Utrecht, The Netherlands) and the Control Systems group of the Technische Universiteit Eindhoven. FROG provided testing facilities including two omni-directional AGVs. Industrial AGVs are custom made for the transportation tasks at hand and come in a variety of forms. To reduce development times it is desirable to follow a model-based control design approach as this allows generalization to a broad class of vehicles. We have adopted rigid body modeling techniques from the ??eld of robotic manipulators to derive the equations of motion for the AGVs and load in a systematic way. These models are based on physical considerations such as Newton's second law and the positions and dimensions of the wheels, sensors, and actuators. Special emphasis is put on the modeling of the wheel-??oor interaction, for which we have adopted tire models that stem from the ??eld of vehicle dynamics. The resulting models have a clear physical interpretation and capture a large class of vehicles with arbitrary wheel con??gurations. This ensures us that the controllers, which are based on these models, are applicable to a broad class of vehicles. An important prerequisite for achieving smooth cooperative behavior is that the individual AGVs operate at the required accuracy. The performance of an individual AGV is directly related to the precision of the estimates for the odometric parameters, i.e. the effective wheel diameters and the offsets of the encoders that measure the steering angles of the wheels. Cooperative transportation applications will typically require AGVs that are highly maneuverable, which means that all the wheels of an individual AGV ahould be able to steer. Since there will be more than one steering angle encoder, the identi??cation of the odometric parameters is substantially more dif??cult for these omni-directional AGVs than for the mobile wheeled robots that are commonly seen in literature and laboratory settings. In this thesis we present a novel procedure for simultaneously estimating effective wheel diameters and steering angle encoder offsets by driving several pure circle segments. The validity of the tuning procedure is con??rmed by experiments with the two omni-directional test vehicles with varying loads. An interesting result is that the effective wheel diameters of the rubber wheels of our AGVs increase with increasing load. A crucial aspect in all control designs is the reconstruction of the to-be-controlled variables from measurement data. Our to-be-controlled variables are the planar motion of the load and the motions of the AGVs with respect to the load, which have to be reconstruct from the odometric sensor information. The odometric sensor information consists of the drive encoder and steering encoder readings. We analyzed the observability of an individual AGV and proved that it is theoretically possible to reconstruct its complete motion from the odometric measurements. Due to practical considerations, we pursued a more pragmatic least-squares based observer design. We show that the least-squares based motion estimate is independent of the coordinate system that is being used. The motion estimator was subsequently analyzed in a stochastic setting. The relation between the motion estimator and the estimated velocity of an arbitrary point on the vehicle was explored. We derived how the covariance of the velocity estimate of an arbitrary point on the vehicle is related to the covariance of the motion estimate. We proved that there is one unique point on the vehicle for which the covariance of the estimated velocity is minimal. Next, we investigated how the local motion estimates of the individual AGVs can be combined to yield one global estimate. When the load and AGVs are rigidly interconnected, it suf??ces that each AGVs broadcasts its local motion estimate and receives the estimates of the other AGVs. When the load is semi-rigidly interconnected to the AGVs, e.g. by means of revolute or prismatic joints, then generally each AGV needs to broadcasts the corresponding information matrix as well. We showed that the information matrix remains constant when the load is connected to the AGV with a revolute joint that is mounted at the aforementioned unique point with the smallest velocity estimate covariance. This means that the corresponding AGV does not have to broadcast its information matrix for this special situation. The key issue in the control design for cooperative transportation tasks is that the various AGVs must not counteract each others' actions. The decentralized controller that we derived makes the AGVs track an externally provided planar motion setpoint while minimizing the interconnection forces between the load and the vehicles. Although the control design is applicable to cooperative transportation by multiple AGVs with arbitrary semi-rigid AGV-load interconnections, it is noteworthy that a particularly elegant solution arises when all interconnections are completely rigid. Then the derived local controllers have the same structure as the controllers that are normally used for individual operation. As a result, changing a few parameter settings and providing the AGVs with identical setpoints is all that is required to achieve cooperative behavior on the velocity level for this situation. The observer and controller designs for the case that the AGVs are completely rigidly interconnected to the load were successfully implemented on the two test vehicles. Experi ments were carried out with and without a load that consisted of a pallet with 300 kg pave stones. The results were reproducible and illustrated the practical validity of the observer and controller designs. There were no substantial drawbacks when the local observers used only their local sensor information, which means that our setup can also operate satisfactory when the velocity estimates are not shared with the other vehicles.
Parameter tuning and cooperative control for automated guided vehicles
2005-01-01
Stouten , B 2005 , ' Parameter tuning and cooperative control for automated guided vehicles ' , Doctor of Philosophy , Electrical Engineering , Eindhoven . https://doi.org/10.6100/IR590723
Hochschulschrift
Elektronische Ressource
Englisch
DDC: | 629 |
Towards Cooperative Transportation of Multiple Mecanum-Wheeled Automated Guided Vehicles
British Library Conference Proceedings | 2019
|