ETH Zurich Robotics and Perception Group

ETH Zurich Robotics and Perception Group
Name	Robotics and Perception Group
Affiliation	ETH Zurich
Established	2010
Director	Vladlen Koltun
Location	Zurich, Switzerland

ETH Zurich Robotics and Perception Group

The Robotics and Perception Group is a research laboratory at ETH Zurich focused on autonomous systems, computer vision, and robotic perception. The group conducts experimental and theoretical work that connects sensor processing, machine learning, and control for mobile robots, aerial platforms, and industrial manipulators. It engages with partners across academia and industry to translate advances into applications in autonomous driving, aerial inspection, and augmented reality.

Overview

The group was founded amid developments in visual SLAM, deep learning, and aerial robotics, drawing upon influences from Stanford University, Massachusetts Institute of Technology, Carnegie Mellon University, University of Oxford, and Imperial College London. Its agenda spans perception pipelines, state estimation, and motion planning with connections to milestones such as the DARPA Grand Challenge, the DARPA Robotics Challenge, and competitions like the RoboCup and VSS series. Members publish in venues including IEEE International Conference on Robotics and Automation, IEEE/RSJ International Conference on Intelligent Robots and Systems, Conference on Computer Vision and Pattern Recognition, and Neural Information Processing Systems.

Research Areas

Research synthesizes techniques from sensor fusion, learning-based perception, and control theory. Key themes include visual-inertial odometry informed by techniques from Simultaneous localization and mapping, mapping technologies related to Google Street View, and scene understanding inspired by work from Facebook AI Research and DeepMind. Projects explore monocular and stereo vision with benchmarks tied to datasets such as KITTI, Cityscapes, TUM RGB-D Dataset, and ImageNet. The group investigates neural architectures influenced by ResNet, U-Net, Transformer (machine learning), and probabilistic estimation methods akin to the Kalman filter and Particle filter. Application domains intersect with Autonomous vehicles, Unmanned aerial vehicle, Augmented reality, and Robotic manipulation.

Projects and Collaborations

The group leads and contributes to collaborative projects with industrial partners and consortia including Siemens, ABB Group, NVIDIA, Microsoft Research, and Amazon Robotics. It participates in EU research frameworks and projects analogous to Horizon 2020 initiatives, and collaborates with Swiss institutions such as ETH Board, University of Zurich, and Swiss Federal Institute of Technology Lausanne. Notable efforts include field experiments reminiscent of deployments by Waymo, inspection pilots similar to those by Shell and BP, and perception stacks comparable to prototypes from Tesla. The lab engages in open-science practices by releasing datasets and tools used by communities around OpenCV, ROS, PCL (Point Cloud Library), and repositories connected to GitHub.

People and Organization

The group comprises principal investigators, postdoctoral researchers, doctoral candidates, and technical staff with backgrounds from institutions such as Princeton University, Harvard University, California Institute of Technology, ETH Zurich, and Ecole Polytechnique Fédérale de Lausanne. Leadership and alumni have moved into roles at organizations including Google, Apple, Uber, NVIDIA Research, Boston Dynamics, and startups spun out to commercialize perception technologies. Collaborators and advisors include researchers who have published alongside teams from University of Cambridge, University of Toronto, University of Pennsylvania, Johns Hopkins University, and Cornell University. Graduate students often participate in summer programs and internships hosted by NASA, European Space Agency, and technology centers such as Facebook AI Research.

Facilities and Resources

Laboratory infrastructure supports multi-sensor platforms, motion-capture arenas, and compute clusters leveraging hardware from NVIDIA, Intel, AMD, and cloud services like Google Cloud Platform and Amazon Web Services. Experimental equipment includes quadrotor fleets resembling systems used by DJI, robotic arms comparable to products from Universal Robots and KUKA, and laser scanning devices akin to offerings from Leica Geosystems and Velodyne. The group utilizes middleware and frameworks such as Robot Operating System, simulation tools inspired by Gazebo (software), and visualization systems with lineage to RViz. Dedicated testbeds support long-range mapping projects, high-rate visual-inertial research, and human-robot interaction trials similar to demonstrations at conferences like IEEE VR and ACM SIGGRAPH.

Awards and Impact

Members and alumni have received recognitions and awards linked to communities including the IEEE Robotics and Automation Society, European Robotics Forum, and academic prizes comparable to the ACM Doctoral Dissertation Award and national science foundations. Publications have influenced standards and benchmarks used by teams in competitions such as DARPA Subterranean Challenge and international challenges hosted by MICCAI and CVPR. The group's data releases and open-source software have been adopted by research groups at University of California, Berkeley, University of Michigan, ETH Zurich, and industrial labs at Google Research and Microsoft Research, contributing to advancements in mapping, localization, and visual perception.

Category:ETH Zurich