Visual Navigation Systems

Transcrição

Visual Navigation Systems
Outdoor navigation based on visual
information
WOLF, Denis; OSORIO, Fernando; SIMOES, Eduardo1
PESSIN, Gustavo; FERNANDES, Leandro1
HATA, Alberto; SHINZATO, Patrick; DIAS, Mauricio; COUTO, Leandro1
1
Instituto de Ciências Matemáticas e de Computação - ICMC
Mobile Robotics Laboratory - LRM
Universidade de São Paulo – USP
São Carlos-SP, Brasil
The LRM Lab activities are focused in the development of research and applications using
intelligent mobile robots and vehicles. One of our main research focus is on autonomous
navigation in outdoor environments. In order to navigate safely, robots must be able detect
obstacles and depression, and avoid them. LRM has been working on terrain mapping
techniques that are capable of represent details of the terrain in a 3D representation and
classify the terrain according to its navigability.
Laboratory:
LRM-ICMC-USP
EESC-USP
Web:
INCT- SEC GT2: http://www.inct-sec.org/
LRM-ICMC: http://www.lrm.icmc.usp.br/
Funding:
USP
INCT-SEC
FAPESP, CNPq, CAPES,
Ministério da Ciência e Tecnologia
Computer Assisted Vehicle Driving
LRM: Mobile Robotics Lab. – ICMC - USP São Carlos
Science Facilities
Researchers: Prof. Dr. Denis Wolf, Prof. Dr. Fernando Osório, Prof. Dr. Eduardo Simões
Graduate Students: Gustavo Pessin, Leandro Fernandes (Ph.D. Students)
Alberto Hata, Patrick Shinzato, Mauricio A. Dias, Leandro Couto (M.Sc. Students)
Project SENA: Mechatronics – EESC - USP São Carlos
Researchers: Prof. Dr. Glauco Caurin, Prof. Dr. Marcelo Becker, Prof. Dr. Valdir Grassi Jr.,
Prof. Dr. Adriano Siqueira, Prof. Dr. Daniel Varela Magalhães
General Description:
The LRM Lab (Laboratório de Robótica Móvel / Mobile Robotics Laboratory) is part of the ICMC-USP
(Univesity of São Paulo), located at São Carlos, Brazil. This research laboratory has facilities used to
develop hardware and software for mobile robotic systems, where the LRM activities are aligned and
focused mainly on both three main Task Groups (GT) of the INCT-SEC:
• GT1 – Indoor Tactical Mobile Robot (Service and Security Robots / Single or Multiple Robots)
“Desenvolvimento de robôs táticos para ambientes internos” – Coord. by Prof. F. Osório
• GT2 – Autonomous Grounded Vehicles (Assisted and Autonomous Vehicle Driving)
“Desenvolvimento de veículos terrestres e autônomos” – Coord. by Prof. D. Wolf
• GT3 – Unmanned Aerial Vehicles (Autonomous Airplanes, Aerial Missions)
“Desenvolvimento de Sistemas Aéreos Não Tripulados” – Coord. by Prof. Onofre Trindade Jr.
The LRM Lab activities are focused in the development of research and applications using intelligent
mobile robots and vehicles. One of our main research focus is on autonomous navigation in outdoor
environments.
Outdoor space navigation in real scenarios and unknown terrain are certainly the more complex
problems. Beyond the obstacles avoidance, is necessary that the mobile robot can identify the region
where it can navigate safely. The terrain on outdoors is composed by a variety of elements like grass,
gardens, sidewalks, streets and gravels. These elements may have different colors and textures allowing
developing a vision-based system. The first step to build a vision-based outdoor navigation system is to
classify outdoor spaces into two classes: navigable region and non-navigable region. The navigable
region is the surface where a mobile robot can travel safely. After its detection, other algorithms can
perform path planning and reactive obstacle avoidance.
LRM has been using data obtained by a video camera to identify parts of the terrain that are adequate for
navigation (part of the road). After the image is obtained, several attributes are calculated for each part
of the image (block). After that, we use a supervised learning algorithm to classify each image block
into navigable or non-navigable.
Science Facilities
Terrain classification: green areas are correctly classified
Our experiments have been conducted in realistic environments under hard situations such as: dirty
roads and different light conditions. The more recent version of the classification algorithm is based on a
combination of artificial neural networks.
Confidence level
Road detected
Multiple ANN image classification.
Science Facilities
Additional References:
1) Shinzato, P. Y. ; Wolf, D. F. . Path Recognition for Outdoor Navigation. In: IEEE Latin American Robotics Symposium,
2009, Valparaiso. IEEE Latin American Robotics Symposium, 2009. p. 1-5.
2) Shinzato, P. Y. ; FERNANDES, L. C. ; OSORIO, F. S. ; Wolf, D. F. . Path Recognition for Outdoor Navigantion Using
Artificial Neural Networks: Case Study. In: nternational Conference on Industrial Technology, 2009, Vina del Mar. IEEE
International Conference on Industrial Technology - ICIT 2010, 2009. p. 1-6.