[Novelty Detection with Robots Using the Grow-When-Required Neural Network]
Date: Sep 2018
Technologies used: C++ 11, QT 5, CMake, ARGoS simulator, Bash scripts, Git
The Grow-When-Required Neural Network (GWRNN) implementation for novelty detection in simulated robot experiments using the ARGoS robot simulator.
The GWRNN [1,2] is capable of learning an input space representation in a self-organised, unsupervised fashion and detect when a novel input is presented. The network runs inside a robot that travels through the environment, progressively learns its features and recognises when a novel object is placed into the environment.
This code repository includes:
The simulation code varies the network, robot and environmental parameters and enables precise control of the variability of the environment and of the noise in the robot sensors. Data from the experiments is stored in text files with tab-separated values, organised in folders specified for each experiment. These text files can be analysed using my custom data analysis code.
[2] Neto, H. V., & Nehmzow, U. (2007). Real-time automated visual inspection using mobile robots. Journal of Intelligent and Robotic Systems, 49(3), 293–307.
The Grow-When-Required Neural Network (GWRNN) implementation for novelty detection in simulated robot experiments using the ARGoS robot simulator.
The GWRNN [1,2] is capable of learning an input space representation in a self-organised, unsupervised fashion and detect when a novel input is presented. The network runs inside a robot that travels through the environment, progressively learns its features and recognises when a novel object is placed into the environment.
This code repository includes:
- A library for the robot controller that allows the robot to move through an environment and perceive coloured lights
- A class that implements the neural network and that is instantiated inside the robot program
- Libraries that control various environments
- Bash scripts that automatically setup and run large sets of experiments
The simulation code varies the network, robot and environmental parameters and enables precise control of the variability of the environment and of the noise in the robot sensors. Data from the experiments is stored in text files with tab-separated values, organised in folders specified for each experiment. These text files can be analysed using my custom data analysis code.
References
[1] Marsland, S., Shapiro, J., & Nehmzow, U. (2002). A self-organising network that grows when required. Neural Networks, 15(8–9), 1041–1058.[2] Neto, H. V., & Nehmzow, U. (2007). Real-time automated visual inspection using mobile robots. Journal of Intelligent and Robotic Systems, 49(3), 293–307.
{Please enable JavaScript in order to post comments}