0000002740 20W 6SWS PR Masterpraktikum - Cloud-Based Machine Learning in Robotics (IN0012, IN2106, IN4287)   Hilfe Logo

LV - Detailansicht

Wichtigste Meldungen anzeigenMeldungsfenster schließen
Allgemeine Angaben
Masterpraktikum - Cloud-Based Machine Learning in Robotics (IN0012, IN2106, IN4287) 
practical training
Winter semester 2020/21
... alle LV-Personen
Informatics 6 - Chair of Robotics, Artificial Intelligence and Real-time Systems (Prof. Knoll)
(Contact information)
Allocations: 1 
Angaben zur Abhaltung
PLEASE NOTE: Prospective participants must have the required previous knowledge stated below and apply for the course as explained in "Course Criteria & Registration."

Please check the "Additional Information" section to download the slides from the preliminary meeting.

Applying state-of-the-art deep learning methods to robotics remains challenging. Generating the required amount of data for training on physical robots is costly and usually takes too long to be practically feasible. This is why simulation environments are becoming increasingly important in machine learning. They not only reduce costs and setup times but also enable arbitrary acceleration of the learning process through massively parallel deployment in the cloud. However, many state-of-the-art simulation environments only support very simple robot systems are not tailored to the specific requirements in robotics.

The goal of this practical course is to set up virtual environments in a cloud-based robot simulation environment and to train machine learning models in the cloud. Participants will leverage modern deplyoment methods such as containers to deploy detailed simulation models in the cloud and train simulated robot to perform tasks based on simulated sensor input.
Required Knowledge:
- Good proficience in Python
- Machine learning and neural networks (e.g. IN2064)
- Common machine learning tools (e.g. TensorFlow, PyTorch etc.)
- Robotics (e.g. IN2067)
- Basic understanding of C/C++
- Working with Linux, especially on the command line

Beneficial Knowledge:
- Experience with ROS and robot simulations (e.g. Gazebo)
- Basic knowledge in 3D modeling
- Docker
Participants will implement machine learning problems as robot simulations (e.g. pick-and-place tasks, navigation etc.), deploy them on common cloud infrastructures (e.g. OpenStack, Amazon Web Services, Microsoft Azure etc.) and train them using state-of-the-art machine learning models.
  • English
  • German
Für die Anmeldung zur Teilnahme müssen Sie sich in TUMonline als Studierende*r identifizieren.
Note: The registration for the course will be managed by the TUM matching system. Applicants are required to send a brief description of their skills and practical programming experience together with a short paragraph outlining their motivation to join the course and a transcript of records to florian.walter@tum.de by July 23, 2020. An introductory meeting will be held on Zoom (Meeting ID: 999 1355 2312, Password: 012871) on July 13, 2020 from 10:00 - 11:00 am.
Online information
e-learning course (moodle)