Periodic Reporting for period 1 - DEEPFIELD (DeepField- Deep Learning in Field Robotics: from conceptualization towards implementation)
Reporting period: 2019-10-01 to 2020-12-31
INESC TEC is strongly committed to become a centre of excellence with focus on field robotics in the aerial and underwater robotics domain. In the last years, the centre for Robotics and Autonomous systems (CRAS) of INESC TEC has advanced its scientific knowledge in sensing and perception methods for robot’s navigation and localization in harsh operational environments. The key objective of INESC TEC is to become one of the research centres of reference in the European Union in field robotics and, at the same time, contribute to bring robotics technology to solve real life problems with very limited or without human intervention.
Industry 4.0 artificial intelligence (AI), and big-data issues are key European and national priorities. The robotics market is growing exponentially and will continue to do so for the next decade, by having robots capable of performing evermore difficult tasks. One of the scientific topics that is helping widespread robot technology is deep learning. Looking at the main robotics and computer vision scientific conferences, journals, and workshops it is possible to observe a pattern, which is the rapid pace at which deep learning approaches are growing when compared to other conventional state-of-the-art approaches.
However deep learning for robotics applications still poses many unanswered questions at the robotics community, such as how much trust can we put in the predictions of a deep learning system when misclassifications can have catastrophic consequences? How can we estimate the uncertainty in a deep network’s predictions and how can we fuse these predictions with prior knowledge and other sensors in a probabilistic framework? How well does deep learning perform in realistic unconstrained open set scenarios where objects of unknown class and appearance are regularly encountered?
These are scientific topics INESC TEC wants to address in its field robotics applications. Therefore, gaining competencies and knowledge in deep learning is a key success factor to increase INESC TEC group research profile (specially for young researchers), and put INESC TEC in the robotics vision forefront.
Therefore this project aims at creating solid knowledge and productive links in the topic of deep-learning in field robotics between INESC TEC and established leading research European institutions, capable of enhancing the scientific and technological capacity of INESC TEC and linked institutions (as well as the capacity of partnering institutions involved in the twinning widespread coordinating and supporting action (CSA)), helping raising its staff research profile and its recognition as an European research centre of excellence and reference in field robotics.
Industry 4.0 artificial intelligence (AI), and big-data issues are key European and national priorities. The robotics market is growing exponentially and will continue to do so for the next decade, by having robots capable of performing evermore difficult tasks. One of the scientific topics that is helping widespread robot technology is deep learning. Looking at the main robotics and computer vision scientific conferences, journals, and workshops it is possible to observe a pattern, which is the rapid pace at which deep learning approaches are growing when compared to other conventional state-of-the-art approaches.
However deep learning for robotics applications still poses many unanswered questions at the robotics community, such as how much trust can we put in the predictions of a deep learning system when misclassifications can have catastrophic consequences? How can we estimate the uncertainty in a deep network’s predictions and how can we fuse these predictions with prior knowledge and other sensors in a probabilistic framework? How well does deep learning perform in realistic unconstrained open set scenarios where objects of unknown class and appearance are regularly encountered?
These are scientific topics INESC TEC wants to address in its field robotics applications. Therefore, gaining competencies and knowledge in deep learning is a key success factor to increase INESC TEC group research profile (specially for young researchers), and put INESC TEC in the robotics vision forefront.
Therefore this project aims at creating solid knowledge and productive links in the topic of deep-learning in field robotics between INESC TEC and established leading research European institutions, capable of enhancing the scientific and technological capacity of INESC TEC and linked institutions (as well as the capacity of partnering institutions involved in the twinning widespread coordinating and supporting action (CSA)), helping raising its staff research profile and its recognition as an European research centre of excellence and reference in field robotics.
The Project activities will be split according with the different work packages: Project management (WP1), Raising Research Profile (WP2), Reinforcement of Scientific and Technological Potential (WP3), Young researchers support and knowledge leverage (WP4) and Dissemination and Communication (WP5). The three-year project will be divided in 4 major periods (9 month each) where each partner will be in charge for one of the challenging issues in deep-learning and field robotics. Each period will be composed by a research formation set composed by a 1 thematic workshop, 1 short-term scientific mission and 1 hands-on summer/winter school.
Due to the COVID19 Pandemic and associated constraining measurements, only one thematic workshop and one short term scientific mission was achieved. Workshops, Summer/Winter schools and Short Term Scientific Missions are all activities that require travelling which was prohibited since March of 2020.
Due to the COVID19 Pandemic and associated constraining measurements, only one thematic workshop and one short term scientific mission was achieved. Workshops, Summer/Winter schools and Short Term Scientific Missions are all activities that require travelling which was prohibited since March of 2020.
With the mass vaccination for COVID19 in progress, travelling will probably be permitted in the end of 2020 and therefore presential activities will be restored. Nevertheless Thematic Workshops will be held in a virtual format in order to reduce travelling and to ensure the accomplishment of the activity without compromising the apprenticeship. The second half of the project will accommodate all activities that were postponed, resulting in a busier schedule.
Young researchers will be encouraged to participate in international events and to publish their work in peer reviewed journals. The teams network will gradually increase as well as the number and quality of projects submitted.
Young researchers will be encouraged to participate in international events and to publish their work in peer reviewed journals. The teams network will gradually increase as well as the number and quality of projects submitted.