Research in artificial intelligence (AI) and machine learning (ML) often requires going from development and testing on personal single GPU workstations to large scale experiments on massive data centers. The scalable AIOps platform, provided at Linköping University, is uniquely suitable for such purposes and highly relevant to most AI, ML, and robotics research.
Based on many years of experience working in several different AI/ML frameworks, the research group of Fredrik Heintz has developed a light-weight DevOps platform for AI and machine learning. The platform builds upon custom made tools mainly developed from open-source software. It supports containerization, detailed version control on multiple levels, and thereby reproducibility.
The platform greatly simplifies and speeds up research by allowing faster and better experimentation, especially when the methods transition from local development to execution on large-scale computing infrastructure. It works equally well for AI/ML and robotics research including highly complex full AI-robotics stacks.
For robotic research the DevOps platform includes version-controlled development environment, build environment, and run-time environment inside docker/singularity containers. The entire robot stack, as well as virtual world simulators and virtual sensors (e.g., camera, LIDAR, IMU) run inside containers that can be readily deployed on any real/virtual machine.
The platform can provide support to many groups within ELLIIT. It supports a wide range of use-cases from simpler ML applications using data sets, to interactive reinforcement learning applications, to complex robotic systems involving many different components. A research engineer is available to support researchers and provide software support, as well as accelerate and broaden the scope of machine learning by allowing large and complex experiments to run on the platform.
Fredrik Heintz (Linköping University)
Fredrik Lindsten (Linköping University)