Home - gestom/stroll GitHub Wiki
Spatio Temporal Representations for Life-Long Mobile Robot Navigation
STRoLL is a research project aimed to create affordable robots capable of reliable long-term operation in outdoor, naturally-changing environments. In the end of this 3-year project, our robots will be able to understand the natural processes that change the environment appearance over time and use this knowledge to improve their ability to self-localise and navigate using only off-the-shelf, inexpensive sensors.
Main objectives
The project has three main objectives: Development of spatio-temporal representations that will allow mobile robots to learn from their experience and understand the naturally-occurring changes in their environment. Development of the visual navigation, localization and mapping methods that will create, update and exploit the spatio-temporal models so that the robot will be capable of long-term autonomous navigation. Exhaustive dataset collection that will allow to benchmark the spatio-temporal models and navigation methods at the given testing sites throughout the entire project duration.
Previous work
The STRoLL project will build upon the previous research aimed at robust visual navigation in environments that exhibit environment changes (see GRIEF, of SurfNav) and the concept of Frequency Map Enhancement FreMEn developed in the STRANDS project.
Goals achieved
- Improvement and software release of a teach-and-repeat vision navigation system
- Extension of the Frequency Map Enhancement to non-binary states - Warped hypertime models
Additional releases
- Day-night dataset relase of Hostibejk site.
- ROS driver for the MMP-5 robot
- ROS driver for the ECA Cameleon robot
Project team
- Dr. Tomas Krajnik, principal investigator, http://labe.felk.cvut.cz/~tkrajnik
- Tomas Vintr, data mining for inference of spatio-temporal representations.
- Filip Majer, computer-vision methods robust to environment changes.
- Lucie Halodova, building and maintenance of feature-based maps for long-term navigation.