| Seznam |
| Téma: | Rychle uceni a rizeni realnich robotu pomoci neuronovych siti |
|---|---|
| Katedra: | Katedra kybernetiky |
| Vedoucí: | Ing. Teymur Azayev |
| Vypsáno jako: | Diplomová práce, Semestrální projekt |
| Popis: | The project is an active area of research in the deep learning community.
You will be working with a real platform (RC buggy or Quadcopter or other platform if you have) to collect data and learn a model (forward function) which will be used for learning a control policy "in simulation" which will then be tested on the real platform. Since we typically have few data points from a real model, an available meta-learning algorithm will be used to learn the function from fewer samples. Part of the task will involve the use of ensembles or other techniques which provide confidence of learned model at various points. The robot platforms are simply designed from a raspberry pi 4 and a servo control board. The platforms are mostly or fully ready for use. |