Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Object-Level Semantic Mapping in Challenging Environments
The goal of this project is to investigate strategies that can help increase the robustness of instance-level mapping pipelines, particularly focusing on aerial or construction robotics use cases.
Keywords: Object Detection and Segmentation, Geometric-Semantic Data Fusion, 3D Mapping
Sensing and modeling their surroundings with high fidelity is a fundamental requirement for mobile robots. While geometric maps allow robots to perform basic tasks such as navigation, higher-level semantic information is often key to increase the richness with which robots can understand the world around them and interact with the objects or humans in it.
Driven by the recent advances in deep learning, semantic mapping, i.e. the process of attaching a semantic label to the entities being mapped, has become a very active research topic in computer vision and robotics during the past few years. Some of the ongoing research directions in the lab involve topics such as dynamic object handling, multi-view geometric-semantic data fusion and semantically informed active mapping. Particularly, we focus on the application of such methods in challenging environments, e.g. for aerial or construction robotics use cases. Students interested in any of these areas are encouraged to contact the main supervisor to arrange an individual discussion about potential projects.
Sensing and modeling their surroundings with high fidelity is a fundamental requirement for mobile robots. While geometric maps allow robots to perform basic tasks such as navigation, higher-level semantic information is often key to increase the richness with which robots can understand the world around them and interact with the objects or humans in it. Driven by the recent advances in deep learning, semantic mapping, i.e. the process of attaching a semantic label to the entities being mapped, has become a very active research topic in computer vision and robotics during the past few years. Some of the ongoing research directions in the lab involve topics such as dynamic object handling, multi-view geometric-semantic data fusion and semantically informed active mapping. Particularly, we focus on the application of such methods in challenging environments, e.g. for aerial or construction robotics use cases. Students interested in any of these areas are encouraged to contact the main supervisor to arrange an individual discussion about potential projects.
- Literature research on online instance/semantic mapping methods.
- Design and implementation of an instance/semantic mapping pipeline.
- Experiments with simulated and real data.
- Evaluation of the proposed approach and comparison against the state of the art.
- Literature research on online instance/semantic mapping methods. - Design and implementation of an instance/semantic mapping pipeline. - Experiments with simulated and real data. - Evaluation of the proposed approach and comparison against the state of the art.
- Background in computer vision and deep learning.
- Python and/or C++ programming experience.
- Familiarity with Linux and ROS is beneficial.
- Background in computer vision and deep learning. - Python and/or C++ programming experience. - Familiarity with Linux and ROS is beneficial.