Career
After postdoctoral research at ETH Zurich and the GRASP Laboratory at the University of Pennsylvania, where he worked with Vijay Kumar and Kostas Daniilidis, Scaramuzza joined the University of Zurich in 2012 as an assistant professor. [6] He became a tenured professor in 2017 and founded the Master's program in Artificial Intelligence at UZH in 2021. He held visiting positions at Stanford University and has collaborated extensively with research institutions including MIT, Caltech, Max Planck Institute for Intelligent Systems, and INRIA. In 2025, he was appointed Distinguished Visiting Scientist at NASA's Jet Propulsion Laboratory, contributing to autonomous vision systems for the Cadre and Endurance lunar missions and the Mars Science Helicopter.
Scaramuzza's research has appeared in The New York Times , [7] BBC News , [8] [9] [10] la Repubblica , [11] and Neue Zürcher Zeitung , [12] MIT Technology Review , [13] Wired , [14] [15] and IEEE Spectrum . [16] [17] [18]
Research
Event-based vision for low-latency perception
Scaramuzza is regarded as one of the pioneers of event-based vision, a neuromorphic sensing paradigm for event cameras, in which pixels respond asynchronously to brightness changes rather than recording full image frames. [25] His group introduced the first algorithms for event-based visual odometry, corner detection, tracking, and SLAM. [26] Later work developed asynchronous convolutional and graph neural networks for scene understanding with sub-millisecond latency. [27]
Applications from his lab include drones that recover from motor failure by relying on event-based sensing [28] , dodging fast moving objects with reaction times of 3.5 milliseconds [29] , and 5,000 Hz detection of traffic participants at the bandwidth of a 50 Hz standard camera. [30]
Scaramuzza's group also created foundational resources such as the ESIM event camera simulator and annotated datasets for autonomous driving, such as DSEC. [31] The field has since grown into a mainstream area at CVPR, ICCV, and ECCV, and has influenced major industrial investments by Sony, Samsung, Huawei, Apple, Meta, and OmniVision. In 2021, Sony and Samsung released their first commercial event cameras, and in 2023 the European Space Agency deployed them in orbit. His contributions in this domain earned the Misha Mahowald Prize, an ERC Consolidator Grant, the IEEE Kiyo Tomiyasu Award, and elevation to IEEE Fellow. [32]
Super-human agile flight
Another strand of Scaramuzza's work investigates vision-based navigation for agile quadrotor flight. His group demonstrated the first autonomous drones to perform acrobatic maneuvers, navigate forests and snowy terrain up to 40 km/h, and race against professional human pilots. [33]
In 2022, they reported in Nature the first system capable of defeating world champions in drone racing, using only an onboard camera and reinforcement learning policies trained in simulation with minimal fine-tuning in the real world. This marked one of the first demonstrations of an autonomous robot achieving world-champion-level performance in a physical sport, comparable to milestones such as AlphaGo or AlphaStar in games. [33]
Follow-up work introduced abstractions such as optical flow maps and trajectory-level outputs to bridge the simulation-to-reality gap, allowing drones to generalize across environments and hardware platforms. [34] Other studies compared reinforcement learning against optimal control methods, showing the former's advantages in robustness and adaptability. [35]
Vision-based navigation of micro drones
In 2009, Scaramuzza led a team that built the first quadrotor capable of fully autonomous flight using only an onboard camera and IMU, winning the European Micro Aerial Vehicle Competition autonomy track. [36]
He subsequently developed the Semi-Direct Visual Odometry (SVO) algorithm, combining direct and feature-based methods for real-time visual SLAM. SVO ran at 100 Hz onboard a 250 g quadcopter in 2014 and later became core IP for Zurich Eye, acquired by Meta Platforms to develop the Oculus Quest headset. SVO has been adopted in products by DJI, Magic Leap, Huawei, Nikon, Parrot, and Hilti. [37]
His group also introduced on-manifold preintegration techniques for visual–inertial odometry, which reduced computational costs and became standard in commercial SLAM systems. [38] Further contributions established the research area of perception-aware planning, where quadrotor trajectories are optimized to maintain visual feature observations. This enabled agile maneuvers such as flying through narrow gaps, perching on power lines, and racing through cluttered environments. [39] [40]
This page is based on this
Wikipedia article Text is available under the
CC BY-SA 4.0 license; additional terms may apply.
Images, videos and audio are available under their respective licenses.
