ESP | CAT | ENG

A new technique developed at the UdG improves global positioning systems


Josep Aulinas has developed a new technique based on independent submaps and a global map. This global map contains the relative transformations between the submaps, which are updated when previously known zones are revisited. They are then fused to maintain the correlations between the vehicle and the landmarks. Using submaps reduces the calculation cost and improves the reliability of the final map.

Mobile robots need global positioning systems to perform their functions completely autonomously. GPS is a good example of that. The GPS signal is not always accessible, or it isn’t available, and on sometimes it is erroneous due to perturbations. On those occasions something more is needed.

Josep Aulinas of the VICOROB research group of the University of Girona (UdG) carries out research in simultaneous localisation and mapping (SLAM).

These techniques use the information from sensors on board a vehicle to generate a map of the area through which the device is moving, while at the same time the vehicle’s position on the map is estimated. SLAM algorithms don’t produce accurate maps of large areas because uncertainty gradually increases on long-term missions. As the map grows larger the computation cost increases, so SLAM solutions cannot be applied in real-time.

In his doctoral thesis Josep Aulinas presents a study of SLAM techniques, with special emphasis on those meant to be used in large areas. He has also studied SLAM applications in underwater environments.

Innovation
The new technique, called Selective Submap Joining SLAM (SSJS) for Autonomous Vehicles, is based on the use of independent submaps and a global stochastic map.

The global map contains the relative transformations between maps, which are updated when previously known zones are revisited. Landmarks already on the map are observed and the submaps sharing information are fused, maintaining the correlations between the vehicle and the landmarks. The use of submaps reduces the calculation cost and improves the reliability of the map.

A viable and less expensive system
A database of reference within the SLAM community, the Victoria Park data set, was used to compare the proposed technique to existing techniques. Synthetic and experimental results show that the SSJS is capable of reliably generating maps for large areas at a lower cost than other techniques.

Once the viability of the system is demonstrated, the method can be adapted for use in autonomous underwater vehicles (AUV). In his thesis, which is titled Selective Submap Joining SLAM for Autonomous Vehicles, Josep Aulinas presents two series of experiments: one with the REMUS-100 AUV equipped with a side scan sonar, and the other with the SPARUS AUV equipped with a downward looking camera. The SPARUS is a vehicle designed and developed entirely by members of the VICOROB group.
Under the supervision of his directors Joaquim Salvi and Yvan Petillot, Josep Aulinas has developed artificial vision techniques to extract and match landmarks required for the SLAM algorithms. The experiments carried out using real data demonstrate the system’s capacity to localise the vehicle and generate maps, combining automatic landmark extraction with the SSJS.


Puntuación
1 Star2 Stars3 Stars4 Stars5 Stars
no hay comentarios Síguenos en facebook y en twitter

Proyecto financiado por el Ministerio de Educación, Cultura y Deporte en el marco del Programa Campus de Excelencia Internacional

Universitat de les Illes Balears ⋅ Ctra. de Valldemossa, km 7.5 ⋅ Palma ⋅ 07122 Illes Balears ⋅ Tel: +34 971 173 000
Universitat de Girona - Campus e-MTA ⋅ Parc Científic i Tecnològic, Edifici Casademont ⋅ C. Pic de Peguera, 15, ⋅ 17003 Girona Tel. 972 419 678
CSIC. Serrano, 117 ⋅ 28006 Madrid ⋅ Tel: +34 914 113 077
Institut Català de Recerca de l’Aigua (ICRA)Edifici H2O - Parc Científic i Tecnològic de la Universitat de Girona. Emili Grahit 101- 17003 Girona

2012 © e-MTA Campus Euromediterráneo del Turismo y el Agua AVISO LEGAL| MAPA WEB
iglesiesassociats