Part 7/10
The task of pouring a liquid into a cup requires precise motion control of the robot and adaptability to varying conditions, such as different liquid levels, cup positions and cup shapes. Conventional methods reach their limits here, as they are often inflexible and difficult to adapt to new situations. Machine learning methods offer a promising alternative.
With machine learning, the system can learn by interacting with its environment, develop optimal strategies and adapt dynamically to changing conditions. For HIRAC we rely on current machine learning methods, in particular deep learning and reinforcement learning.
Reinforcement Learning (RL)
Reinforcement learning is a field of machine learning in which a so-called agent learns to make optimal decisions by interacting with its environment.
Through success and failure in training, the agent tries to develop a strategy that minimises failures. The criteria for success and failure were defined in advance together with Offenburg University of Applied Sciences and serve as the basis for training.
Machine learning plays such an important role in this project because the system is to be extended to different robot arms. With reinforcement learning, the learnt model can be transferred to different robots and retrained with minimal adjustments. After a short training period, the system can, for example, learn different motion sequences for different robot arms.
kununu