- aroefer [at] @ [at] cs.uni-freiburg.de
- ARoefer
- aroefer.de
- Robot Learning Lab
Adrian Röfer, M.Sc.
I am a PhD student working at the Robot Learning Lab at the University of Freiburg and also a member of the ELLIS Society's network of researchers. My interests lie with all kinds of robotic manipulation, but my main focus has been on developing methods to enable Manipulation with a Purpose - manipulation that is motivated and tightly coupled with the constraints and goal of the task a robot is supposed to solve. I have worked on multiple projects in service of this interest, starting with student projects during my undergraduate and graduate degree, and continuing with two theses in this field. As my previous efforts have been model-based I setting out to utilize model-free learning methods in service of this interest as a part of my PhD research.
During my graduate degree, I collaborated frequently with the Laboratory for Progress at the University of Michigan. As a part of these collaborations, I explored generating motion controllers from semantic predicates which were grounded in symbolic mathematical descriptions. This exploration later inspired the topic of my graduate thesis.
Further, I collaborated with Dr. Zhen Zeng on the paper "Semantic Linking Maps for Active Visual Object Search". The paper was published at ICRA 2020 and received the IEEE ICRA Best Paper Award in Cognitive Robotics.
On practical note, I was able to apply my experience in robotics to a real-world scenario during an internship at the Amazon Development Center in Berlin. Different from my academic work, I focused on achieving roboustness in the given problem space, as opposed to exploring a particular method across multiple spaces.
Further, I got to test and improve my software development skillset while working as an SD in projects with the FFT Produktionssysteme GmbH & Co. KG. While this work was not robotics related, it introduced me to the world of embedded development and enabled me to focus on organizing and creating production-level software.
Curriculum Vitae
Below you can browse through my CV.
Publications Education Work ProjectsPublications
Kineverse: A Symbolic Articulation Model Framework for Model-Generic Software for Mobile Manipulation
Adrian Röfer, Georg Bartels, Wolfram Burgard Abhinav Valada, and Michael Beetz.
In: IEEE Robotics and Automation Letters, 2022
Robot Skill Adaptation via Soft Actor-Critic Gaussian Mixture Models
Iman Nematollahi, Erick Rosete-Beas, Adrian Röfer, Tim Welschehold, Abhinav Valada, Wolfram Burgard.
In: IEEE International Conference on Robotics and Automation (ICRA), Philadelphia, USA, 2022
Doing Right by Not Doing Wrong in Human-Robot Collaboration
Laura Londoño, Adrian Röfer, Tim Welschehold, Abhinav Valada.
In: ACM/IEEE International Conference on Human-Robot Interaction Workshop on Fairness and Transparency in Human-Robot Interaction, 2022
RobotIO: A Python Library for Robot Manipulation Experiments
Lukas Hermann, Max Argus, Adrian Röfer, Thomas Brox, Abhinav Valada, Wolfram Burgard.
In: arXiv preprint arXiv:2207.13591, 2022
Semantic Linking Maps for Active Visual Object Search
Zhen Zeng, Adrian Röfer, and Odest Chadwicke Jenkins.
In: IEEE International Conference on Robotics and Automation (ICRA), 2020
Education
PhD
University of Freiburg
2021 - present
Learning long-horizon mobile robotic manipulation from demonstration, semantic perception, and language supervision.
Disseration: TBD
MSc. Computer Science
University of Bremen
2017 - 2020
Focus on robotics and artificial intelligence, specifically robotic manipulation.
Thesis: Kineverse: A Framework for Maintaining and Exchanging Articulation Models for Mobile Manipulation Robots.
BSc. Computer Science
University of Bremen
2013 - 2017
Focus on robotics and artificial intelligence, specifically robotic manipulation.
Thesis: A Comparison of Two Control Schemes for Obstacle Avoidance in Robotic Shelf Replenishment Tasks.
Work Experience
PhD Student
University of Freiburg
June 2021 - Present
PhD Student at the Robot Learning lab headed by Prof. Abhinav Valada. Researching mobile manipulation, visuo-motor control, and learning from demonstrations.
Software Developer
Aero Test Europe GmbH
December 2020 - January 2021
Software developer at Aero Test Europe. Working in a joint project with FFT Produktionssysteme GmbH & Co. KG. investigating opportunities for utilizing IoT devices in testing of aviation systems. Development of infrastructure enabling the communication of larger systems with application specific embedded devices.
Contractor
University of Bremen
August 2020
Contractor tasked with integrating the Kineverse articulation model framework with the Giskard motion planning library being developed at the Institute for Artificial Intelligence (IAI) at the University of Bremen.
Robotics Intern
Amazon Development Center Berlin GmbH
November 2019 - March 2020
Internship at Amazon Development Center in Berlin. Researched improvements to industrial robotic manipulation tasks. Developed low-level robot controller for optimal task performance. Implemented a new method of motion generation for the application in collaboration with an academic consultant. To enable a data-driven development process, created multiple tools for recording and processing performance data, and for predicting future system performance from historic data.
Research Collaborator
University of Michigan
August 2018 - September 2018
Second author on a research paper to be submitted to IROS 2019. The paper contributes to the field of indirect object search in large scale environments. Symbolic object relations are combined with metric observations to derive hypothetical object locations, guiding the search. The search itself uses a tightly integrated perception-control loop, allowing for dynamic camera control and occlusion avoidance.
Research Assistant
University of Bremen
April 2018 - July 2018
Research assistant at the Institute for Artificial Intelligence headed by Prof. Michael Beetz, tasked with evaluating the PyBullet physics engine's simulation capabilities regarding soft bodies and force-torque sensors. Developed an object oriented, easy-to-use wrapper for the engine and built a lightweight simulator on it for fast prototyping of motion controllers.
Research Intern
University of Michigan
October 2017 - February 2018
Research internship at the Laboratory for Progress headed by Prof. Chadwicke Jenkins. Helped with the first implementation of the giskardpy motion control library. Implemented a lightweight velocity control interface and giskardpy interface for the Fetch robot. Worked on a project investigating the automatic generation of task-space motion controllers based on background knowledge. The project aimed at integrating predicates for state assessments and motion controllers tightly, allowing for a direct generation of motion controllers from predicates. The resulting system was used in the joint research project mentioned above.
Visual Artist
Private Client
2017
Created digital set extensions and did compositing work for a stop motion film using the 3D creation suite Blender.
Research Assistant
University of Bremen
April 2017 - July 2017
Research assistant at the Institute for Artificial Intelligence, tasked with creating a development environment for the giskard motion control framework. The environment was implemented as RVIZ plugin and provides basic features supporting fast controller development and testing.
Teaching Assistant
University of Bremen
October 2014 - September 2016
Teaching assistant for the courses Computer Science and Media I/II. Gave weekly tutorial sessions on the lecture's current topic and graded students' assignments.
Contractor
University of Bremen
2014
Created a tutorial on the basics of using the 3D creation suite Blender for modeling and animating of 3D geometry, as well as using Blender for camera tracking and compositing.
Project Experience
University of Bremen
October 2016 - September 2017
Students' robotics project SUTURO at the Institute for Artificial Intelligence. Goal of the project was to build a robotic café at the institutes's laboratory. Customers would be able to put in orders, which would then be prepared and delivered by a robot. Designed motion controllers for the PR2 robot, giving it needed manipulation capabilities. Developed new programming languages, features and debugging tools for the giskard motion control framework to ease development and testing of motion controllers.
University of Bremen
April 2016 - July 2016
4-person class project for lecture Entertainment Computing. Goal of the project was to plan and prototype a video game using the Unity game engine. Contributed to the initial concept, implemented game logic in C# and created all visual assets for the prototype.
University of Bremen
October 2015 - July 2016
Students' robotics project SUTURO at the Institute for Artificial Intelligence. Goal of the project was to solve an object retrieval task involving multiple objects and a tray in the institutes's laboratory. The PR2 robot would be able to autonomously locate and retrieve various items in the environment. Designed manipulation capabilities for the PR2 robot using the MoveIt! library.
University of Bremen
October 2014 - March 2015
6-person software project part of the class Software Project II. Goal of the project was to create a software tool for a local elementary school, helping the staff with creation and management of schedules. Implemented database interface and GUI in Java and JavaFX for the tool.