Surveyor Robotics Journal

Surveyor Robotics Journal

	email: support@surveyor.com Surveyor Corporation home page Subscribe to a syndicated RSS feed. Archives November 2008 October 2008 September 2008 August 2008 September 2008 August 2008 July 2008 June 2008 May 2008 April 2008 March 2008 February 2008 January 2008 December 2007 November 2007 October 2007 September 2007 August 2007 July 2007 June 2007 May 2007 April 2007 March 2007 February 2007 January 2007 December 2006 November 2006 October 2006 September 2006 August 2006 July 2006 June 2006 May 2006 April 2006 March 2006 Februray 2006 January 2006	Thu, 08 Feb 2007 2007 AAAI Spring Symposium - Robots and Robot Venues: Resources for AI Education Just wanted to note that we signed on as a sponsor for the 2007 AAAI Spring Symposium on Robotics at Stanford University - March 26-28. We'll be there with a bunch of robots, so if you are planning to attend the conference, please look us up. Here are the specific details - http://www.cs.hmc.edu/roboteducation/ Posted Thu, 08 Feb 2007 18:50 \| HTML Link \| see additional stories ... "Evolutionary Robotics" and the SRV-1 It's been a month since I first posted about our exploration of adaptive / emergent behaviors for the SRV-1. In the interim, we have experimented with a variety of neural network architectures and reviewed a number of research studies. It has been an interesting journey, as we have since abandoned several of our original assumptions, e.g.: 1. that we would want create some basic neural net functions for color and feature classification that could be employed by our onboard procedural code in C or BASIC or offboard control by Python, MSRS, etc 2. that we needed to create an internal topological representation of the outside world for purpose of localization In fact, we have found that neural network architectures can be central to the operation of the robot, employing no procedural code whatsoever, instead directly connecting to processed sensor inputs and motor outputs, and that neural models can effectively model the topology of the "real world" without explicit coding of those features in structured internal representations. The most interesting and useful studies to support this revised viewpoint are found in the MIT Press publication (2000) *"Evolutionary Robotics - The Biology, Intelligence, and Technology of Self-Organizing Machines"* by Stefano Nolfi and Dario Floreano, as well as in subsequent published works by the authors. There are now numerous researchers employing this approach to robotics, as shown on The Evolutionary Robotics Homepage From a practical perspective, we are moving forward on this new path, specifically working on the task of training a soccer robot, as first described here, by defining a sequence of robot training tasks with progressively increasing complexity, e.g.: robot locates and pushes ball robot locates and pushes ball in bounded area robot locates and pushes ball in bounded area with obstacles robot locates and pushes ball toward goal in bounded area robot locates and pushes ball toward goal in bounded area with obstacles We will create, in SRV-1 firmware, an artificial neural net "brain" whose architecture can be defined by the user, with inputs supplied by core image processing routines (e.g. scan() and blob()), and outputs going direct to the motors. We plan to provide two modes of learning - supervised training, and unsupervised evolutionary self-organization. The unsupervised mode will require the use of a simulator to explore multiple generations of evolution in a reasonable timeframe, but we have a good plan for this (to be discussed in a future post). The real goal is to make the process of non-procedural robot learning accessible and understandable to any user, and this is an exciting challenge ! Posted Thu, 08 Feb 2007 13:22 \| HTML Link \| see additional stories ...