Refine
Year of publication
- 2022 (2)
Document Type
Language
- English (2)
Has Fulltext
- no (2)
Keywords
Institute
Is part of the Bibliography
- yes (2)
The handling of flexible components, such as rubber sealings or cables, is an essential part of industrial manufacturing. Automating such processes through robotics is still a major challenge. Machine learning-based training of robots in combination with simulation has led to promising results in other areas of robotics. This approach can potentially be applied to assembly tasks as well. However, current simulation tools for robotics generally do not have capabilities for simulating flexible bodies. In this work, an approach for modeling such flexible bodies is proposed, which can be implemented with most common robotic simulation programs. Using the example of a rubber O-ring sealing, this approach was tested in such program and the results were compared with a professional multi-body physics engine. Realistic behavior of the O-ring was achieved with relatively low computation times.
Comparison of Commercial Physics-Based Simulation Environments with a Collision-Rich Benchmark
(2022)
For many years simulations are an important tool for researchers in many fields of application. With ongoing improvements in computing power and the rising requirements in the industry, simulation software has to shift the focus more towards industrial use cases. Especially in robotics and machine learning applications, the physical behavior has to be reproduced as realistic as possible to avoid unwanted collisions during runtime. In general, applications such as mounting or bin picking with an automated robot shall be evaluated. Therefore this paper investigates three in-use and one newly published software environment with included physics engine for realistic simulations that are commercially available. For comparison, a benchmark with collisions between two gears to imitate a mounting process is implemented and key features are defined. Only two of the simulation software manage to pass the benchmark. Despite that, both of them show restrictions revealing a lack of reliable software for realistic robot interaction simulation.
