Advanced Control

Using sensors and actuators is the fundamental principle of robotics – and the processing between stimulus and response the intelligence that defines the robot. The general term Cybernetics describes the trans-disciplinary approach governing the control of machines, organisms and organizations. In robotics we basically distinguish between open-loop and closed-loop (feedback) control systems. Whereas open-loop control is not capturing its output, in closed-loop control the output is continuously fed back through sensor measurement.

INTRODUCTION

Cybernetics - the art of control

Therefore, feedback control is used in robotics on various abstraction levels enabling the execution and pursuit of targeted actions. More sophisticated control approaches are able to adapt to changing constraints and may learn optimal control strategies autonomously. From a practical point of view, control theory is deeply interconnected with communication because the delay of a control loop has influence on the control quality, e.g. dynamic positioning accuracy of electrical drives.

We provide tailored control solutions for all levels of robot control ranging from low-level control of electric drives to adaptive and iterative learning process control based on in-process quality measurements. One focus of our work is visual control using different camera technologies enabling e.g. the grasping of moving parts or mobile navigation.

We at GESTALT have experts in different fields of control for robotics and we are working with clients from different sectors on projects tackling innovative control solutions, taking into account real-time constraints, sensor integration and respective communication technologies

TECHNOLOGICAL OUTLINE

The (numeric) control chain and abstraction levels of control

The numeric control is transforming control commands into working and movement procedures of robots. Therefore, control is the basis for explicit robot programming by the user as well as for fully autonomous behavior. Abstraction of the general concept of control in the direction of the robot hardware or otherwise in the direction of the overall application is considered low-level or high-level control. The different degrees of abstraction are also referred to as control levels, which can be fundamentally differentiated into product, process or task, (end-effector) movement and axis control. Low-level control typically relies on hard real-time requirements in order to guarantee accuracy for position and velocity. Hence, the lower the degree of abstraction of the control functions, the more accurate and faster the behavior of the robot has to be controlled. In principle, high-level control is more efficient and requires less specific expertise. In addition, the design of high-level control functions is closer to natural human-human communication and control, but increases demands on the sensory and cognitive capabilities of the robotic system.

On the lowest level of industrial robot control, the robot behavior is controlled by giving input by numeric axis variables for the individual axes. On the motion level, movements of the robot flange or end-effector are defined by Cartesian coordinates. At the process level, the specification of a sequence of sub-operations is given. On the product level, only the desired properties of the workpiece are given as input. The robot can abstract these properties autonomously, transfer them into actions and rules and execute them.

APPLICATIONS & USE CASES

Bin-picking and Visual Servoing – Flexible gripping of (moving) parts

Classic tasks in industrial robotics presume a known and structured environment. The robots were controlled through predefined sequences of poses. This behavior is typically optimized towards low cycle times and high productivity at the cost of flexibility. Typically these kinds of robot are not able to adapt towards changing environmental conditions. Modern demands in production and service robotics call for a high degree of flexibility. The challenge is to understand unstructured environments in order to process varying work-piece geometries, positions, manipulation sequences and more.

Bin-Picking is a term describing the challenge to pick up work-pieces piled at random from pallets or boxes. This challenge is solved with the help of cameras and image processing involved in the control loop by object pose estimation, calculation of gripping point and specific gripping technologies. There are different strategies for attaching the camera either to the robot tool following the “Eye-in-hand” principle or with a fixed camera in the workspace. Whereas bin-picking can typically be solved with an open control loop, the challenge gets harder if one needs to pick up moving objects.

A closed-loop approach called Visual Servoing uses visual feedback extracted from a camera sensor to control the motion of the robot. Position-based visual servoing is continuously controlling the pose of the end-effector in regard to a control deviation (difference between the current pose and the target pose). The object pose is calculated based on images (2D or 3D) and offset against the current TCP pose. The pose error is used to determine a Cartesian control following and approaching the object until it is available for gripping.

*Position-based Visual Servoing principle*

CONCLUSION

From basic motion control to learning control algorithms and intelligent process adaption

Control is the fundamental basis for all robotic applications. The principle connects sensory input and actuator actions. This applies to industrial manipulators as well as to mobile service robots. We consider different control levels with varying requirements for real-time capabilities of processing and communication. Furthermore we target different approaches towards additional sensor integration in favor of flexibility. We have plenty of experience in designing control functions and control systems for machines and processes. This includes building real-time robot control systems from scratch as well as developing and integrating tailored visual or learning control functions. Get in touch with us to discuss your use case. You can contact us under info@gestalt-robotics.com or give us a call at +49 30 616 515 60 – we would love to hear from you.

Stefan3. September 2019