Ocean Systems Lab  
Ocean Systems Lab Heriot-Watt University Ocean Systems Laboratory

Seminar Information

Title: Behaviour Recognition for Spatially Unconstrained Unmanned Vehicles
Acoustic source localization using the Adaptive Projected Subgradient Method
Avoidance Method for mobile robot, using the DVZ
Date: Wed 24 Jun 2009
Time: 11:00
Location: EM 2.33
Speaker: Rolf Baxter
Nicolas Valeyrie
Pierre Regis
Description: =================================================
Behaviour Recognition for Spatially Unconstrained Unmanned Vehicles
Rolf Baxter

We use trajectory based techniques to perform location independent behaviour recognition on an unmanned underwater vehicle. Unmanned vehicles have applications in both surveillance and structural inspection, but require robust, location independent behaviour recognition. Previous research has used GPS or location based states that tie model parameters to specific locations, and have rarely considered performance under specific levels of noise. We use location independent action based states to recognise high level behaviour using flat Hidden Markov Models. We validate this approach by comparing performance under different levels of noise, achieving 78% classification precision under 50% corruption.

======================================================
Acoustic source localization using the Adaptive Projected Subgradient Method
Nicolas Valeyrie

The presentation is about localizing an acoustic source using the Adaptive Projected Subgradient Method. This method uses a network of nodes distributed in the area of interest and working in the diffusion mode of cooperation. The nodes can communicate with their neighbours and thereby improve their own estimate of the acoustic source position. The use of the Adaptive Projected Subgradient Method allows the topology of the network to change as well as the transmissions between the nodes to temporarily fail. Under some hypothesis, there is asymptotically consensus, that is the nodes have the same estimate of the source position.

===================================
Avoidance Method for mobile robot, using the DVZ
Pierre Regis

In this presentation, I will show a quite simple algorithm for obstacle avoidance in creating a Deformable Virtual Zone surrounding the robot, using external sensors. The global idea is to give a command to the robot in order to recreate the initial shape of the Virtual Zone when it is deformed by an obstacle.

<< Back to list of seminars

  © 2008-2017 HW Ocean Systems Laboratory