Part 6/10
So far, most AIs in robotics have been used to recognise, grasp and precisely position objects. Using the object to fulfil a particular purpose has not yet been sufficiently researched. Our software must therefore be tested regularly and trained with artificial intelligence during the development phase.
To test the software and optimise the algorithms, the system is trained with a mixture of real hardware and simulation. For the simulation we use specific software that can simulate motion sequences in robotics. The simulation software can be used to test the use of different sensor systems or robotic arms, and deviations can be examined and taken into account in software development.
In addition to the simulation test, a realistic environment is being developed so that the AI-based software can be tested and trained. The environment is recorded under different lighting conditions in order to train the system for use at different times of day and in different places. Possible operating problems are also identified through tests with test subjects at Offenburg University of Applied Sciences.
These tests allow problems to be identified and eliminated from the outset. In the process, the algorithms are adjusted and the artificial intelligence is fed with the information obtained. In this way, tasks are to be performed identically in the simulation environment and in the real world, despite different hardware.
kununu