Student Projects

To apply, please send your CV, your Ms and Bs transcripts by email to all the contacts indicated below the project description. Do not apply on SiROP Since Prof. Davide Scaramuzza is affiliated with ETH, there is no organizational overhead for ETH students. Custom projects are occasionally available. If you would like to do a project with us but could not find an advertized project that suits you, please contact Prof. Davide Scaramuzza directly to ask for a tailored project (sdavide at ifi.uzh.ch).

Upon successful completion of a project in our lab, students may also have the opportunity to get an internship at one of our numerous industrial and academic partners worldwide (e.g., NASA/JPL, University of Pennsylvania, UCLA, MIT, Stanford, ...).

  • MPC for high speed trajectory tracking

    More …

  • In this project we want to explore the possibility of active sensing using an event camera, where the task is to do asynchronous 3D reconstruction.

    More …

  • Develop a planning framework for very fast or even time-optimal quadrotor trajectories.

    More …

  • Develop an embedded low level flight controller for quadrotors on existing hardware.

    More …

  • The goal of this project is explore ways to adapt existing deep learning algorithms to handle sparse asynchronous data from events.

    More …

  • The goal of this project is to explore new algorithms for processing events within a deep-learning context.

    More …

  • Work on decentralized SLAM that requires a minimum of data exchange between robots. Focus will be on Map Optimization / Bundle Adjustment

    More …

  • Exploit sparsity of data flow in spiking neural network simulations

    More …

  • An exploration of optimization methods for spiking neural networks

    More …

  • Use machine learning to condense visual observations of places into minimal representations that allow for 6dof relative pose estimation.

    More …

  • Learn depth from RGB frames and sparse depth information.

    More …

  • Teach and repeat, but try to be as fast as possible on repeat. Goal is to deploy this on a quad.

    More …

  • Build a Neural Network that takes as input an image, a sequence of previous images, and the geometry extracted from that, and produces as output a descriptor which can be used for place recognition.

    More …

  • Try to handle failures in april/aruco tag detection using deep learning.

    More …

  • Train a neural network to detect poses of arbitrary objects

    More …

  • Building a simulator that combines the photorealistic image rendering engine with the ROS framework could greatly help the robotics research community to develop algorithms with the simulator. For example, both two popular open-source simulators, CARLA and AirSim, are supporting ROS.

    More …

  • Using photorealistic game engine, such as Unity or Unreal Engine, for vision-based AI research has become increasingly popular in the robotics community since it is faster to generate and automatically annotate high-quality synthetic image data. Existing drone simulators, such as RotorS or Flightgoggles, are either not providing high-quality synthetic images or not supporting accurate quadrotor dynamics.

    More …

  • Visual-Inertial Odometry is a great solution for drone-navigation in GPS-denied environments. Its ability to provide centimeter-level precision in local navigation makes it a suitable choice in many commercial applications like last centimeter drone delivery. Conventional VIO algorithms work well in static environments. However, when the environment is dynamic, i.e most of the visual features come from a moving environment, for instance a moving platform, the VIO does not perform reliably. This problem can be attributed to the unreliable initialisation phase of the VIO pipeline, which is the most critical phase. Most initialisation algorithms are based on structure-from-motion, which assumes that the environment is static. In such a scenario the initialisation algorithm needs to be adapted to take into account the motion of the features.

    More …

  • Dense stereo matching is crucial for creating depth maps. However, it is very computationally expensive on CPUs. As a result the update rate is low, which makes it unusable for many tasks, eg. avoidance in dynamic environments. Especially on drones the computation power is limited. Through the introduction of the Nvidia Jetson series computers we got access to lightweight embedded GPUs. Running the stereo matching on a GPU can potentially make it significantly faster.

    More …

  • VIO system are used to navigate drones in GPS denied environments. These systems are, however, prone to failure, for example due to loss of feature tracking. This will result in a crash, as the drone has no knowledge about its own state. However, if the drone manges to hold its position without the VIO system for some time, the VIO pipeline can be reinitialised and the navigation task can continue. Additionally, it will also prevent a crash, increasing the safety. The images from the camera and IMU data are usually still available when the VIO pipeline crashes.

    More …

  • In order for UAVs to fly and especially land autonomously we need evaluate the safety of landing spots. Having semantic information about the scenery, eg. road, forest, roof, etc, enables the UAV to pick a safe landing spot.

    More …

  • This project focuses on the object tracking, (i.e., target following) on nano-UAVs (few centimeters in size).

    More …

  • In this project, we aim to build a self-supervised depth estimation and segmentation algorithm by embedding classic computer vision principles (e.g. brightness constancy) into a neural network.

    More …

  • The project aims to develop techniques based on machine learning to have maximal knowledge transfer between simulated and real world on a navigation task.

    More …

  • Explore machine learning based approaches for deblurring of images with events

    More …

  • Investigate the usability of data-driven methods to improve the performance of a VIO pipeline on a resource-constrained platform.

    More …

  • Combine state-of-the-art place embedding and feature extraction networks into a single network and train them jointly.

    More …

  • Achieve comparable or better place recognition performance than NetVLAD, with a simpler network architecture.

    More …

  • This work investigates the usage of Gaussian Processes for uncertainty-aware system identification of a quadrotor platform.

    More …