Open Master’s or Honor’s Thesis

Thesis: Teaching a Robot to High-Five

robotIn this project the student will use a modern machine learning method -Interaction Primitives- to implement human-robot interaction. The goal is to teach a humanoid robot how to perform a high-five and how to anticipate a high-five from a human partner. To this end, electromyography (EMG) data will be used to predict the final location of the humans movement. The robot will then move his arm towards this position. The task of the student will be to (1) apply and refine interaction primitives, (2) analyze and predict EMG data, and (3) perform robot experiments. Ideally, the project concludes with a final experiment/demonstration on one of the robots. Main intended learning outputs for the student are:

  • Learn about modern machine learning methods
  • Work with state-of-the-art humanoid robot
  • Learn how to predict EMG data

Required skills are

  • Knowledge in Matlab and Python
  • Linux
Research Implementation Technical Robotics Mathematical / Machine Learning
rating5 rating5 rating3 rating3 rating3

Summer Projects 2016

The summer projects presented below are open for student volunteers for summer 2015. While those projects cannot be funded, i.e. no research assistant or similar positions are available, credit hours can be acquired. If you are interested in one of the projects or if you have any questions, fell free to contact us via email or through the contact form at the end of this page.

Robotic Infrastructure Project

all_robots

Goal of this project is to improve and setup parts of the infrastructure of the Intelligent Robotics Lab. The lab has currently three to four different types of robots: A Baxter, an UR 5 and several Robotis manipulators and robots. Most of the robots make use of the Robotics Operating System (ROS) and two different simulators are used.

The result of this summer project should be several interfaces for Matlab and C++ or Python for each robot, incorporating the different sensors and capabilities. The interfaces are encapsulating the communication to the robotic systems over network or to the simulators, and are going to be used in research and teaching. Some first implementations exist already.

Ideally, the project concludes with a final experiment/demonstration on one of the robots. Main intended learning outputs for the student are:

  • Get familiar with ROS
  • Collect (first) experience with real robots
  • Experience the difference between simulation and real execution

Required skills are

  • Knowledge in Matlab and Python
  • Linux
  • Taking responsibility and care with expensive equipment
Research Implementation Technical Robotics Mathematical / Machine Learning
rating1 rating5 rating3 rating5 rating0

 Deep Learning for Robotics


There is a possibility to present the results of this project during an Nvidia workshop at ASU in June!


 

jetson-tx1-developer-kits In this project the student is designing and executing an experiment using a Deep Learning library on a robotic application. We have for this project two new NVIDIA Jetson TX1 boards combining a CPU and GPU core.

What is Deep Learning? End-to-End_Deep_Learning_Diagram Deep Learning describes basically the use of a neural network with several layers. In the past, neural networks could only be used with a quite limited number of (node-) layers (between an input and output node) but computational advancements and the development of new learning methods increased the number of possible layers.

Do I have to implement deep neural networks by myself? Luckily, several frameworks/libraries already exist for Deep Learning. This project is more about the design and experimental evaluation of such deep neural networks rather than writing an own library. However, some literature research about Deep Learning might be necessary.

Do I have to know Deep Learning for choosing this project? Not necessary, actually Deep Learning is not taught in most of the Computer Science programs. Basic knowledge about machine learning and neural networks (perceptrons, feedforward neural networks) would be sufficient. Otherwise, the willingness to gain the necessary background through (scientific) literature.

Is it worth (the effort) to do something with Deep Learning? Yes, Deep Learning is currently fancy in research and industry as well.

At the end of the project the student should have done an experiment combining Deep Learning and Robotics as well as a documentation and setup of the NVIDIA Jetson boards.

Main intended learning outputs for the student are:

  • Learn about and use Deep Neural Networks
  • Set up and design an own experiment with evaluation
  • Collect experience with NVIDIA Jetson and parallelization

Required skills are

  • Knowledge in Matlab, C++ or Python
  • Linux
Research Implementation Technical Robotics Mathematical / Machine Learning
rating2 rating4 rating1 rating2 rating4

Interested in one of the projects? Contact us now!

Please, let us know which background (taken classes, interest, past work etc.) you have and in which project(s) you are interested.