Motivation
In modern production facilities, gripping robots perform assembly and loading tasks, enabling extensive automation of these processes. However, current gripping systems are usually limited to individual, predefined tasks, meaning that converting or setting up new production lines requires considerable expert knowledge. Increasing product individuality and shorter innovation cycles therefore necessitate a higher degree of autonomy in robotic gripping planning. This enables the development of new industrial sectors by performing complex, non-repetitive gripping movements that would otherwise require human labor. One of these new industrial applications is disassembly, which combines the positive effects of economic efficiency and sustainability in a circular economy.
Overall goal
The goal of the GSAK project is to develop a robust grasp synthesis for industrial robots that explicitly takes downstream actions into account. The central basis for this is affordances, which describe the area of an object that must be grasped in order to perform a downstream task, such as tightening a screw. Another focus is on manipulation in complex, confined environments where targeted contacts and unavoidable collisions with other objects are an integral part of the grasping process. The goal is to significantly increase the gripping success rate for both known objects, such as via CAD models, and similar or novel objects.
Methodology
For a successful grasp synthesis, a method for determining affordances is being developed and coupled with motion planning for the gripper. An experimental and simulated test setup is being created in which an industrial robot can perceive its working area using an 3D camera. This data is combined with an semantic that describe the assembly via an graph in order to train the parameters of the deep learning model. A heuristic extends this data-based process, improves generalization to unknown objects, and ensures an efficient learning process. Both jaw and suction grippers are investigated. Gripping is made more difficult by complex, confined environments in which collisions with other objects must be explicitly included in the gripping plan. The resulting method for grasp synthesis is compared with state-of-the-art methods using benchmarks and a custom industrial dataset for affordances.
Project funding
The GSAK project is funded by the Federal Ministry for Economic Affairs and Energy as part of the Central Innovation Program for SMEs (ZIM).