Competition: AI Olympics with RealAIGym
Organized by the DFKI research departments Robotics Innovation Center and Systems AI for Robot Learning, the "AI Olympics with RealAIGym" competition focuses on the athletic intelligence of robots that allows them to perform highly dynamic tasks. The goal of the competition, which is based on the DFKI initiative called "RealAIGym: Education and Research Platform for Studying Athletic Intelligence", is to advance research in this field and promote the further development of learning and control algorithms. For the participating teams, the initial task was to use an underactuated double pendulum robot system to perform the swing up from a suspended position and its subsequent stabilization in a simulation environment. The teams that mastered this challenge best now have the chance to test their developed algorithms remotely on real hardware in the Underactuated Lab at the DFKI Robotics Innovation Center in Bremen. The results of the competition will be presented by DFKI researchers Dr. Shivesh Kumar and Dr. Boris Belousov at IJCAI 2023. The conference contribution will be complemented by keynotes, among others by Prof. Frank Kirchner (DFKI, Bremen) and Prof. Jan Peters (DFKI, Darmstadt), a presentation about the RealAIGym project, presentations of the best teams and a panel discussion.
Competition: Intrinsic Error Evaluation during Human-Robot-Interaction
Together with the University of Duisburg-Essen (UDE), the Robotics Innovation Center is also hosting the competition "Intrinsic Error Evaluation during Human-Robot Interaction". The idea behind the competition is to use the human electroencephalogram (EEG) to improve human-robot interaction. For the first phase of the contest, the researchers provided participating teams with a partially labeled EEG dataset recorded during experimentation with a robotic elbow orthosis. In the process, the system behaved incorrectly several times, inducing a so-called error potential in the subject's brain. The task of the participating teams was to develop and train a machine learning algorithm based on this data set that is capable of recognizing the error potential and thus the erroneous behavior of the system. In the second part of the challenge, the teams are tasked with proving the performance of their AI model on unlabeled data in real time. To do this, a test subject's EEG data will be measured and streamed live online for all teams to access. The results of the competition will be presented by DFKI researchers at the IJCAI. Keynote speakers will be Prof. Dr. Elsa Kirchner (UDE and DFKI) and Prof. Dr. Frank Kirchner (DFKI, Bremen), among other experts.
Demo Paper: A Human-in-the-Loop Tool for Annotating Passive Acoustic Monitoring Datasets
The DFKI research department Interactive Machine Learning is represented with two scientific contributions at the IJCAI 2023. In the Demonstration Track of the conference program, DFKI scientists Hannes Kath, Prof. Dr. Thiago Gouvêa and Prof. Dr. Daniel Sonntag present their paper "A Human-in-the-Loop Tool for Annotating Passive Acoustic Monitoring Datasets". In it, they describe the demonstration of a software tool that can be used to annotate acoustic data. To this end, the researchers transfer high-dimensional audio data into a 2D space that can be displayed on a computer interface. A model is trained so that similar data points (each representing one second of audio) are located close to each other. In this way, multiple data points can be annotated simultaneously, simplifying the overall annotation process.
Project Proposal: Interactive Machine Learning Solutions for Acoustic Monitoring of Animal Wildlife in Biosphere Reserves
The second scientific contribution from the DFKI research department in Oldenburg, which will be presented within the Special Track on AI and Social Good, introduces a project that aims to facilitate adaptive management of animal biodiversity in wildlife sanctuaries with the aid of interactive machine learning methods. The authors of the paper “Interactive Machine Learning Solutions for Acoustic Monitoring of Animal Wildlife in Biosphere Reserves” include DFKI researchers Prof. Dr. Thiago S. Gouvêa, Hannes Kath, Ilira Troshani, Bengt Lüers, and Prof. Dr. Daniel Sonntag, as well as several international ecologists tasked with management of marine and terrestrial ecosystems in biosphere reserves spread over three continents.
Further information
Website for the IJCAI 2023: https://ijcai-23.org/
Website of the DFKI competitions: https://ijcai-23.dfki-bremen.de/
Scientific contact persons at DFKI
Competition: AI Olympics with RealAIGym
Dr. Shivesh Kumar (Robotics Innovation Center), Dr. Boris Belousov (Systems AI for Robot Learning)
Competition: Intrinsic Error Evaluation during Human-Robot-Interaction
Prof. Dr. Elsa Kirchner (Robotics Innovation Center), Marc Tabie (Robotics Innovation Center)
Demo Paper: A Human-in-the-Loop Tool for Annotating Passive Acoustic Monitoring Datasets
Hannes Kath (Interactive Machine Learning)
Project Proposal: Interactive Machine Learning Solutions for Acoustic Monitoring of Animal Wildlife in Biosphere Reserves
Prof. Dr. Thiago S. Gouvêa (Interactive Machine Learning)