data-science

Bosch Group

<div><section><div><p></p><h2>Company Description</h2></div><div><p><strong>At Bosch, we shape the future by inventing high-quality technologies and services that spark enthusiasm and enrich people’s lives. Our promise to our associates is rock-solid: we grow together, we enjoy our work, and we inspire each other. </strong><strong>Join in and feel the difference.</strong></p><p>The Robert Bosch GmbH is looking forward to your application!</p></div></section><section><div><p></p><h2>Job Description</h2></div><div><p>In SLAM (simultaneous localization and mapping), we consider a mobile robot that creates a map of its environment and determines its own pose relative to that map. Lifelong SLAM considers the case where SLAM is performed over a long period of time, where the same places are revisited many times, and where the environment may be subject to change. In such a scenario it can be necessary to relocalize after a failure in the SLAM algorithm or if the robot is kidnapped, e.g., moved while turned off.</p><p>For this purpose, place recognition algorithms can be employed to find the robot's pose inside a large map, e.g., by finding similar features in the robot's environment and a location in the map. Existing research sometimes used handcrafted features, but in recent years, learned features have become popular. In the field of computer vision, learned features have often outperformed handcrafted image features, and thus, the use of learned features also seems promising for other problems such as lidar SLAM.</p><p>The milestones of the project are as follows.</p><ul><li><strong>Preliminary steps:</strong> Read related literature and get familiar with the tools (ROS, C++, Python, TensorFlow/PyTorch)</li><li><strong>Milestone 1:</strong> Implement and train a neural network using TensorFlow or PyTorch that can learn features based on 2D lidar data.</li><li><strong>Milestone 2:</strong> Adapt the existing C++ relocalization implementation to make use of these features.</li><li><strong>Milestone 3:</strong> Develop an evaluation pipeline to quantify the performance of the learned features in comparison with existing methods such as place recognition algorithms based on handcrafted features as well as machine learning.</li><li><strong>Milestone 4:</strong> Report and final presentation.</li></ul></div></section><section><div><p></p><h2>Qualifications</h2></div><div><ul><li><strong>Education: </strong>studies in the field of Computer Science, Electrical Engineering or comparable</li><li><strong>Personality and Working Practice:</strong> ability to work independently, to carefully implement and evaluate algorithms, and to clearly present the results</li><li><strong>Experience: </strong>in working with mobile robotic applications, localization algorithms, and SLAM, familiar with machine learning algorithms, in particular deep learning</li><li><strong>Know-how:</strong> strong C++ and Python programming skills and good Knowledge of at least one machine learning framework (Tensorflow or PyTorch), experience with ROS</li></ul></div></section><section><div><p></p><h2>Additional Information</h2></div><div><p>If available please send us your Github profile.</p><p><strong>Start:</strong> according to prior agreement<br>
<strong>Duration:</strong> 6 months</p><p>Requirement for this thesis is the enrollment at university. Please attach a motivation letter, your CV, transcript of records, examination regulations and if indicated a valid work and residence permit.</p><p><strong>Need further information about the job?</strong><br>
Gerhard Kurz (Business Department)<br>
<br>
Apply now in just 3 minutes!</p></div></section></div>

Thesis Feature-based Place Recognition Using 2D Lidar Data for Lifelong SLAM

remote

17 days ago