Student Projects

Collision Avoidance - Master Thesis at Avientus

Autonomous Systems Lab

Avientus is a startup that specializes in developing cutting-edge, heavy-duty automated drone transportation systems designed to revolutionize logistics and industrial applications. To further enhance the safety and reliability of their drones, we are offering a Master Thesis opportunity in the field of collision avoidance for drones.

Keywords

Collision avoidance, Computer vision, Drones

Labels

Master Thesis

Description

Requirements

Contact Details

More information

Open this project...

Published since: 2025-08-12 , Earliest start: 2025-08-01 , Latest end: 2026-02-28

Applications limited to ETH Zurich

Organization Autonomous Systems Lab

Hosts Mascaro Rubén , Chli Margarita

Topics Information, Computing and Communication Sciences

Underwater Depth Estimation from RGB and Sonar with Spatio-Temporal Consistency

Autonomous Systems Lab

Are you excited about applying machine learning to real-world robotics challenges? This thesis offers the opportunity to develop a novel depth estimation framework that fuses RGB camera and sonar data for robust underwater perception—crucial in environments where traditional sensors fail. You'll work with real data collected from a Tethys ONE robot and enhance your model with spatio-temporal consistency to boost performance over time. If you're motivated by impactful research, hands-on experimentation, and pushing the boundaries of perception in low-visibility conditions, we’d love to hear from you.

Keywords

robotics, underwater, computer vision, learning, rgb, sonar, depth estimation

Labels

Master Thesis

Description

**Motivation** On land, active light sensors such as LiDAR and RGB-D cameras are widely used, providing metrically accurate and intuitive depth measurements. However, in underwater environments these sensors face significant challenges due to nonlinear and frequency-dependent attenuation, refraction, and scattering caused by suspended particles. While passive RGB cameras can still capture high-resolution information underwater, their effective range is severely limited by turbidity, blurriness, and reflections. Forward Looking Sonars (FLS) offer a complementary sensing modality that is less affected by visibility conditions and provides a greater range, though typically at a lower resolution and loss of spatial information. Combining the strengths of both RGB and FLS sensors to robustly estimate depth is a highly desirable goal. Producing accurate depth maps from such multi-modal data not only enhances perception in low-visibility environments but also enables compatibility with a wide range of existing algorithms that rely on depth as a key input. Furthermore, most learning-based depth estimation approaches treat each input independently (i.e. one-shot inference), disregarding the temporal continuity inherent in sensor data streams. However, in largely static environments, leveraging spatio-temporal consistency across frames can significantly improve robustness and accuracy. **Proposal** We propose the development of a learning-based depth estimation framework that fuses RGB camera and Forward Looking Sonar (FLS) data. To further mitigate the effects of sensor noise and poor visibility, we aim to incorporate spatio-temporal consistency, e.g. by feeding the previous depth estimate back into the model as additional input at each timestep. Given the scarcity of datasets that include RGB, FLS, and ground truth depth data, we adopt a teacher-student learning strategy. Initially, a teacher model will be trained on publicly available underwater datasets containing RGB images and corresponding ground truth depth. Subsequently, we will collect new RGB and FLS data using a Tethys ONE underwater robot. The teacher model will then generate pseudo-labels for this new data, which will be used to train a student model that learns to estimate depth from RGB and FLS inputs. Additionally, photogrammetric reconstruction may be employed to obtain more accurate ground truth depth for portions of the collected dataset, providing a second source of supervision. Finally, we will evaluate the RGB + FLS model against the baseline teacher RGB-only model through both numerical metrics and qualitative inspection. We will also assess the utility of the estimated depth maps for downstream tasks such as mapping and navigation.

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-07-22

Organization Autonomous Systems Lab

Hosts Oleynikova Helen, Dr.

Topics Information, Computing and Communication Sciences , Engineering and Technology

Continuous-Time Multi-Sensor Odometry in the Wild

Autonomous Systems Lab

This project provides the opportunity for an academic exchange with the Vision for Robotics Lab (V4RL) at the University of Cyprus (UCY). Led by Prof. Margarita Chli, the lab has recently won a prestigious ERC grant for the “SkEyes” project, to advance robotic perception for drone swarms and bridging our research activities between ETH Zurich and the University of Cyprus. This master project is offered as part of this research effort.

Keywords

Continuous-Time Odometry, Sensor Fusion

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-07-09 , Earliest start: 2025-07-01 , Latest end: 2026-03-31

Applications limited to ETH Zurich

Organization Autonomous Systems Lab

Hosts Mascaro Rubén

Topics Information, Computing and Communication Sciences

Versatile, Robust and Simulatable Multi-Robot SLAM

Autonomous Systems Lab

Keywords

Multi-Robot SLAM

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-07-07 , Earliest start: 2025-07-01 , Latest end: 2026-03-31

Applications limited to ETH Zurich

Organization Autonomous Systems Lab

Hosts Mascaro Rubén

Topics Information, Computing and Communication Sciences

Bridging Human-Readable and Robot-Perceived Maps: CAD-SLAM Alignment and Refinement

Autonomous Systems Lab

This thesis proposes an industrial collaboration with Sevensense Robotics on enabling robots to take advantage of existing building models for their localization and navigation, by aligning the output of the robot's visual SLAM map to CAD models. This will be an exciting opportunity to push the state of the art in research and also in practical applied demonstrations on real robots.

Labels

Master Thesis , ETH Zurich (ETHZ)

Description

**Background** Sevensense commercializes a state of the art Visual SLAM (Simultaneous Localization and Mapping) system that enables robots to localize and navigate in unknown environments by building 3D maps using visual data from onboard cameras. While these maps are optimized for robotic perception and localization, they often lack the structural clarity and semantic richness of human readable CAD models. This mismatch creates a communication gap between autonomous systems and human operators. Bridging this gap is especially important in industrial settings — such as warehouses or factories — where robots must operate within human-designed infrastructure. Aligning SLAM maps with CAD floorplans not only improves interpretability but also lays the foundation for subsequent tasks like semantic annotation, layout validation, and mission planning. **Goals** The goal of this thesis is to develop and evaluate a novel algorithm for structural alignment which accounts not only for translation and scale but also distortions while maintaining global and local consistency. An initial exploration can be done using linear deformation models which can leverage various correspondences such as corners, walls, or room boundaries. This project is motivated by real industrial use cases at ABB and involves testing on data from deployed AMRs in production settings, making this a challenging scientific problem. As such exploring and testing novel ideas is a core part of the thesis, some of which include: ● Designing non-linear deformation models to project SLAM maps onto CAD structures (e.g., handling distorted corridors) ● Exploring different types of correspondences (e.g., corners, walls, room outlines) to drive the alignment ● Integrating semantic cues or high-level structure ● Developing evaluation metrics for structural alignment quality (e.g., wall straightness, room consistency) ● Leveraging CAD priors to refine noisy or incomplete SLAM reconstructions The approach will be tested on real data collected by ABB’s industrial-grade AMRs, already deployed in environments such as warehouses or production facilities. **Expected outcome** ● A novel spatial registration algorithm to align SLAM-generated maps with architectural CAD models ● A quantitative and qualitative analysis of the proposed method ● Evaluation on real-world SLAM data and deployment on an industrial robot ● Optional extensions: semantic labeling, drift correction, inventory mapping **References** [1] M. Bayón-Gutiérrez, N. Prieto-Fernández, M. T. García-Ordás, J. A. Benítez-Andrades, H. Alaiz-Moretón and G. Grisetti, "CAD2SLAM: Adaptive Projection Between CAD Blueprints and SLAM Maps," in IEEE Robotics and Automation Letters, vol. 10, no. 2, pp. 1529-1536, Feb. 2025, doi: 10.1109/LRA.2024.3522838.

Requirements

Contact Details

More information

Open this project...

Published since: 2025-07-01

Organization Autonomous Systems Lab

Hosts Oleynikova Helen, Dr.

Topics Information, Computing and Communication Sciences , Engineering and Technology

LiDAR-Visual-Inertial Odometry with a Unified Representation

Autonomous Systems Lab

Lidar-Visual-Inertial odometry approaches [1-3] aim to overcome the limitations of the individual sensing modalities by estimating a pose from heterogenous measurements. Lidar-inertial odometry often diverges in environments with degenerate geometric structures and visual-inertial odometry can diverge in environments with uniform texture. Many existing lidar-visual-inertial odometry approaches use independent lidar-inertial and visual-inertial pipelines [2-3] to compute odometry estimates that are combined in a joint optimisation to obtain a single pose estimate. These approaches are able to obtain a robust pose estimate in degenerate environments but often underperform lidar-inertial or visual-inertial methods in non-degenerate scenarios due to the complexity of maintaining and combining odometry estimates from multiple representations.

Keywords

Odometry, SLAM, Sensor Fusion

Labels

Semester Project , Master Thesis

Description

Work Packages

Requirements

Contact Details

More information

Open this project...

Published since: 2025-07-01 , Earliest start: 2025-07-01 , Latest end: 2026-02-28

Applications limited to ETH Zurich

Organization Autonomous Systems Lab

Hosts Mascaro Rubén , Chli Margarita

Topics Information, Computing and Communication Sciences

Note on plagiarism

We would like to suggest every student, irrespective of the type of project (Bachelor, Semester, Master, ...), to make himself/herself familiar with ETH rules regarding plagiarism