Register now After registration you will be able to apply for this opportunity online.
This opportunity is not published. No applications will be accepted.
Bayes Optimal Semantic SLAM using Random Finite Sets
The random finite sets formulation yields algorithms with a high robustness to noise and clutter, especially in the presence of many false positives in the context of feature-based SLAM. In this project, we seek a greater understanding of RFS-based approaches with their application to semantic SLAM.
Keywords: Random Finite Sets, Semantic feature-based SLAM, Semantic information processing, Probabilistic inference
Traditional approaches in feature-based SLAM heavily rely on the data association between measurements and landmarks. Additionally, large measurement noise can significantly downgrade the performance of these approaches. Fortunately, the random finite sets (RFS) formulation yields algorithms with a high robustness to noise and clutter, especially in the presence of many false positives in the context of feature-based SLAM.
Furthermore, as semantic information is mainly unaffected by appearance as well as viewpoint changes, it is a potentially powerful approach for localization, mapping and SLAM in general.
However, efficient and descriptive data representations in the underlying map based on semantics still remain an open research problem.
Recent advances in RFS-based filters such as the generalized labeled multi-Bernoulli filter and the cardinalized PHD filter, enable new state of the art solutions for semantic estimation and mapping.
In this project, we seek a greater understanding of RFS-based approaches with their application to semantic SLAM. Specifically, we would like to combine the advantages of semantic landmarks and RFS models.
Additional information can be found here: https://docs.google.com/document/d/1Zf-MBPpdRcvrsVlK_Ub9pCcJZEGFH_iV1PtqHAj9FJk/edit?usp=sharing
Traditional approaches in feature-based SLAM heavily rely on the data association between measurements and landmarks. Additionally, large measurement noise can significantly downgrade the performance of these approaches. Fortunately, the random finite sets (RFS) formulation yields algorithms with a high robustness to noise and clutter, especially in the presence of many false positives in the context of feature-based SLAM. Furthermore, as semantic information is mainly unaffected by appearance as well as viewpoint changes, it is a potentially powerful approach for localization, mapping and SLAM in general. However, efficient and descriptive data representations in the underlying map based on semantics still remain an open research problem. Recent advances in RFS-based filters such as the generalized labeled multi-Bernoulli filter and the cardinalized PHD filter, enable new state of the art solutions for semantic estimation and mapping. In this project, we seek a greater understanding of RFS-based approaches with their application to semantic SLAM. Specifically, we would like to combine the advantages of semantic landmarks and RFS models.
Additional information can be found here: https://docs.google.com/document/d/1Zf-MBPpdRcvrsVlK_Ub9pCcJZEGFH_iV1PtqHAj9FJk/edit?usp=sharing
- Literature review on finite set statistics and - semantic SLAM
- Implementation of RFS-based filters
- Evaluation of the RFS-based solution
- Literature review on finite set statistics and - semantic SLAM - Implementation of RFS-based filters - Evaluation of the RFS-based solution
- Highly motivated and independent student
- Knowledge in Matlab and/or C/C++
- Strong interest/background in state estimation
- Experience with ROS is beneficial
- Highly motivated and independent student - Knowledge in Matlab and/or C/C++ - Strong interest/background in state estimation - Experience with ROS is beneficial
If you are interested, please send your transcripts and CV to Martin Adams (martin.adams@mavt.ethz.ch) and Lukas Bernreiter (lukas.bernreiter@mavt.ethz.ch).
If you are interested, please send your transcripts and CV to Martin Adams (martin.adams@mavt.ethz.ch) and Lukas Bernreiter (lukas.bernreiter@mavt.ethz.ch).