The future of autonomous gripping
- February 27, 2019
The scope of autonomous gripping technology is expanding beyond rigidly defined operations to embrace the flexibility and autonomy required for the next generation of smart factories.
In the past, says Schunk, gripping processes were primarily geared toward boosting productivity and process reliability. With the advent of smart factories, flexibility is becoming an increasingly important factor. Tomorrow's grippers will enable flexible operations and even autonomous handling scenarios.
Until recently, industrial gripping has been relatively rigid: the geometry of the parts must be known, as well as the exact pick and place position. A reliable handling process can be ensured by predefining traverse paths and specifying target point coordinates based on repeatable parts feeding operations. With the rise of digitalization, the trend is now moving towards highly automated, fully networked and autonomous manufacturing systems.
Against this backdrop, artificial intelligence (AI) is becoming increasingly important. The first cognitive intelligence applications for grippers in combination with cameras are already possible. This allows for intuitive training by the operator and autonomous handling of gripping tasks by the robot. For these applications, Schunk says it designs industry-oriented handling processes by limiting the number of component variations, streamlining the classification and training process.
In an initial use case that makes use of machine learning approaches for workpiece and gripping process classification, interlocking building blocks are randomly combined and presented to a lightweight robot in a random arrangement on a work surface. The robot's task is to pick up and transport the blocks. By interacting with 2D or 3D cameras, the self-learning system rapidly increases gripping reliability after only a few learning cycles. With each grip, the gripper learns how to successfully pick up and transport the workpiece.
Effective learning through continuous optimization
After only a few training sessions, the network classifies how to handle the range of workpieces and the resulting combination options. The gripper knows how to pick up and transport the workpiece based on learned experience. Due to the intelligence of the algorithm, the gripper can classify future combinations and arrangements of workpieces on its own after only a short period of training. In this way, the system is capable to handle parts autonomously and with sensitivity to the situation. The algorithms are continuously adapted using AI methods. This makes it possible to reveal previously unrecognized correlations and further refine the handling process.