A key challenge for robots lies in non-reproducible processes, for example when components are provided in an undefined position or process steps cannot be carried out according to a defined sequence.

Where humans can determine and compensate for the deviation using their acquired process knowledge, the robot must be provided with a model of the process and suitable sensor technology, on the basis of which the control sequence and trajectory can be planned and, if necessary, adapted. Whether a model is required for a specific handling task depends on the requirements and boundary conditions of the task in question. Using the example of a pick-and-place scenario, various options with the corresponding models are listed below.

If components are not stored in a magazine, or more precisely, do not have a known initial position, this information must be provided for the robot controller. For this purpose, kinematic models are used that describe the relationship between positions, speeds and accelerations. If these relationships are known for a component, the question of when the robot end effector should be positioned where and in which pose can be easily answered and the trajectory planned accordingly. A classic industrial example of this is components on a conveyor belt that are provided in a random order. If a robot is to pick them up, its position must be detected. If the conveyor belt is in motion, the belt speed can be taken into account in the gripping point calculation in order to consider the time required to execute the movement.

Problematic momentum of its own

If the dynamic requirements of the component need to be taken into account, a dynamic model is required that also takes the acting forces into account. When handling geometrically large, flat components such as sheet metal, for example, the robot movement can cause a natural oscillation. This increases the force on the gripper, which in the worst case can exceed the gripping force and lead to object loss. It is therefore important to calculate a suitable trajectory in advance using model-based control.

The difficulty with dynamics models often lies in the physical modeling, which quickly becomes complex and time-consuming even for supposedly simple tasks. Further difficulties arise in determining the relevant parameters for the model. If these are too complex to calculate, the real-time requirements of the process may be violated. Furthermore, some of these parameters cannot be identified, which is why model-based control cannot be used.

Simulated object dynamics

It is not only advantageous to consider the dynamics of the component when handling large metal sheets, but also when handling deformable objects such as cables or hoses. These objects are characterized by the fact that their own weight is often sufficient to change not only their position but also their shape. Since a deformation occurs as a result of a force, a dynamic model of the deformation behavior can be used to calculate the shape of the object.

The ISW has developed an approach that takes into account the deformation behavior in the robot's trajectory planning. The manipulated object is modeled as a multi-body system consisting of individual rigid bodies coupled via joints with spring and damper properties. The resulting movement under the force of the robot can then be calculated using a multi-body simulation, whereby the robot is modeled in the simulation alongside the manipulated object. The ISW uses the open-source software DART (Dynamics Animation and Robotics Toolbox) to solve the equation of motion.