Computing Infrastructure
The desktop computing environment of the Department of Empirical Inference is based on Intel PCs, and currently uses centrally managed operating systems Ubuntu Linux, Microsoft Windows, and Mac OS X.
Until recently, storage for research data shares and home directories was provided solely by two department file servers. There is currently a transition toward usage of the NetApp cluster storage operated by the central IT facilities. This solution is shared with the other departments and research groups of the institute. This storage provides hourly snapshot backups. The Empirical Inference owns half (i.e., 120TB) of the data capacity of this cluster. Off-site storage from RZ Garching is offered for long-term archival of scientific data.
The Empirical Inference department used around 32% of the central high performance computing cluster in recent years. Researchers in the department also have access to a powerful supercomputer with a peak performance of more than 20 PFLOPS in the Max Planck Computing and Data Facility in Garching.
Robot Learning Lab: High-Speed Robot Arms
The Robot Learning Lab studies high-speed compliant control for learning tasks such as table tennis, using a maintenance-intensive set-up.
A tendon-driven pneumatic artificial muscle robot arm was developed in-house to study antagonistically actuated joints, allowing for light-weight segments, incorporating strong pneumatic muscles, and using co-contraction for compliant control.
We use this robot to show the beneficial impact of passive compliant muscular actuation in real-world robotics applications.
The software is based on the robot programming framework o80 and makes use of a custom four-camera high-speed vision setup that reliably detects table tennis balls at (200 Hz). A ten camera Vicon motion capture system enables to precisely track objects equipped with markers at 300 Hz. Strong LED lighting, with adjustable intensity and color, support tracking of objects at such frame rates.
For dexterous manipulation tasks, we developed a TriFinger robot setup. The design is focused on robustness and safety, permitting safe operation without human supervision and training reinforcement learning algorithms directly on the real robots.
We have eight such platforms in total. They can be accessed remotely via a job submission system based on HTCondor, thus allowing external collaborators to use the robots as well as the organisation of the "Real Robot Challenge" which enabled safe robotics research even during Covid-19.
Brain-Imaging Equipment
The Brain-Computer Interfaces (BCI) Lab worked in two main areas: Clinical, high-fidelity electroencephalography (EEG) studies, and large-scale EEG studies with lower-fidelity consumer devices.
We have an electromagnetically shielded cabin (mrShield by CFV) for high-fidelity, low-noise laboratory studies. Two 128-channel amplifiers enable high-density EEG, eye (EOG), and muscle (EMG) recordings inside and outside of the laboratory.
For large-scale EEG studies, we have acquired a stock of 34 Muse EEG headbands (InteraXon, Canada). These lower-fidelity consumer devices are ergonomic and easy to use. When paired with a computer or iPad, patients can autonomously participate in large-scale studies at home, with researcher involvement limited to experimental design and data analysis.
Computational Imaging
In computational imaging, it is desirable to develop image processing algorithms that work not just on synthetic but also on real data. To record data for varying setups and with controlled distortions, we use several SLR, CMOS, and CCD cameras with a variety of lenses as well as tele scopes (including a 12" Astro Physics Riccardi-Honders) that allow us to work on real-world astronomical image sequences affected by turbulence.
For quality assessment and quantitative comparisons, we use an image quality analysis system (iQ Analyer by Image Engineering). Additionally, we have developed a precise test-panel that allows us to efficiently measure the properties of imaging optics (such as MTF) in a single shot (see Figure \ref{fig:panel}).