Project: Reconfigurable Ultra-Autonomous Novel Robots

In the general definition of a robot, Sense-Plan-Act, the sense-part require a highly efficient and detailed_x000D_ perception. The most important challenges for autonomous and effective robots in automatic environment_x000D_ recognizing/sensing are the capacity of the robots to see, understand and interact with three-dimensional_x000D_ real world. It is not a coincidence that the visual cortex of the human brain is the largest sensory part of the_x000D_ brain._x000D_ The problem doCO of autonomous robotics comprises of two major tasks with different key aspects. The_x000D_ first one covers exploration issues while creating accurate three dimensional maps. Here, relatively highresolution_x000D_ precise 3D data as well as fast and accurate matching algorithms are required to create consistent_x000D_ scenes. Compare the performance of a human running through a crowed area, an airport, or driving a car._x000D_ Recognizing and identifying an object from a video input turns out to be a very difficult problem. The_x000D_ problem stems from the fact that a single object can be viewed from an infinite number of ways. By rotating,_x000D_ obscuring, or scaling a single object, one can create multiple representations of an object - which makes the_x000D_ problem of matching the object to a database of objects very difficult. Depending on the organization of the_x000D_ database, the problem expands either linearly or exponentially whenever there are numerous objects that_x000D_ should be identified simultaneously._x000D_ The second task covers the exploration and navigation in known and unknown terrains. Real-time 3D_x000D_ computation of the scene in the moving direction of a robot is required to ensure obstacle avoidance,_x000D_ whereas the precision is secondary. The real-time capability is also mandatory for mapping and surveying_x000D_ tasks if environment dynamics are considered. Today, even though cameras provide even more than 30_x000D_ frames per second the existing CPU-based systems cannot execute the necessary cue-extraction and object_x000D_ recognition algorithms, when several cues should be extracted simultaneously, at a rate of more than 5 fps._x000D_ The CO reason for the low rates achieved is the fact that the various cue extraction and object recognition_x000D_ schemes are very CPU intensive tasks; for example it has been reported that robust approaches just for 3D_x000D_ object detection based on stereo processing in dense environment need the performance equivalent of that_x000D_ triggered by more than five high-end CPUs. Obviously, things are getting even more difficult when also_x000D_ considering the CPU load due to the navigation tasks._x000D_ In order to obtain a high performance 3D-perception system a significant amount of parallel computations_x000D_ are required. Given that the availability of image sensors with very high performance at a low cost the_x000D_ challenge is now to create a low cost, high performance Artificial Visual Cortex._x000D_ RUNNER aims at providing a framework based on which highly autonomous Robots with much better_x000D_ perception than the existing solutions, will be created. This innovative infrastructure will utilize state-of-theart_x000D_ reconfigurable devices(FPGAs); Nikitakis, Wyland and Rajan in their papers proved that those devices_x000D_ allow for extremely higher performance and power-efficient processing when implementing data_x000D_ manipulation methods such as 3D sensing/matching schemes as well as template and feature-based object_x000D_ recognition algorithms, while they can be reconfigured on real-time._x000D_ In order to achieve its aims RUNNER will:_x000D_ * Design and implement a family of innovative cue extraction modules supporting very high rates by taking_x000D_ full advantage of the high processing power provided by the high-end FPGAs._x000D_ * Design and implement real-time reconfigurable object sensing mechanisms, which will take advantage of_x000D_ the accurate and fast cue extraction schemes and the high processing power provided by the high-end_x000D_ FPGAs._x000D_ * Design and implement a novel navigation scheme based on the advanced perception provided by the_x000D_ proposed reconfigurable system._x000D_ * Design and implement a sophisticated 3D reconstructing system, tailored to the needs of the cueextraction_x000D_ modules, which will be implemented in FPGAs._x000D_ * Develop and implement the middleware for the seamless programming, configuration and management_x000D_ of the RUNNER infrastructure._x000D_ * Prototype and validate RUNNER’s complete infrastructure and demonstrate its efficiency and wide_x000D_ applicability in two real-world trials._x000D_ The ultimate objective of RUNNER is to deliver a reconfigurable prototype with excessive cross-doCO_x000D_ applicability. In RUNNER, we believe that in a few years there would be millions of robots in various_x000D_ application areas that will all be navigated in an autonomous manner based on 3D video capture; such_x000D_ robots can be efficiently and inexpensively built based on the provided innovative highly flexible_x000D_ infrastructure._x000D_ In order to achieve the above the consortium mobilizes a significant European cross-sectoral force from 5_x000D_ different countries that covers the whole chain of robotics and vision and embedded systems.

Acronym RUNNER (Reference Number: 5527)
Duration 01/12/2010 - 01/05/2014
Project Topic RUNNER aims at providing an innovative infrastructure, to be exploited for the creation of highly autonomous robots. It will utilize high-end reconfigurable devices, in order to allow for extremely high performance and power-efficient processing, when implementing 3D sensing/matching schemes.
Project Results
(after finalisation)
RUNNER project resulted in significant contributions in the field of embedded systems and robotics, in particular in applications in computer vision, object detection, obstacle avoidance and FPGA prototyping and emulation._x000D__x000D_More specificallythe project results included_x000D_ 1. The design and implementation of innovative 3D Reconstruction and Object Detection algorithms and architectures able to deliver accurate perception to the RUNNER vision system _x000D_ 2. Design and development of state-of-the-art Field Programmable Gate Arrays (FPGAs) board_x000D_ 3. Development of initial prototypes of the architectures on the FPGA board_x000D_ 4. Design and development of software tools for synchronization of the different system modules and communication with external sensors and devices _x000D_ 5. Evaluation of the 3D Reconstruction and Object Detection Sub-systems in terms of processing speed, accuracy and hardware overheads_x000D_ 6. Validation of the 3D Reconstruction and Object Detection Sub-systems using real-time data acquisition_x000D_ 7. Validation of the 3D Reconstruction and Object Detection Sub-systems on the RUNNER FPGA Board_x000D_ 8. Validation of the 3D Reconstruction and Object Detection Sub-systems in Robotics Environments_x000D__x000D_
Network Eurostars
Call Eurostars Cut-Off 4

Project partner

Number Name Role Country
9 SignalGeneriX Partner Cyprus
9 University of Cyprus Partner Cyprus
9 Universidad Politecnica de Madrid Partner Spain
9 MEEQ AB (publ) Observer Sweden
9 Ingenieria de Sistemas Intensivos en Software, S.L Partner Spain
9 Mälardalen University Partner Sweden
9 Telecommunication Systems Institute / Technical University of Crete Partner Greece
9 Algosystems S.A. Coordinator Greece
9 Aldebaran Robotics Partner France