ROBOSKIN: Skin-Based Technologies and Capabilities for Safe, Autonomous and Interactive Robots
This project will develop and demonstrate a range of new robot capabilities based on robot skin tactile feedback from large areas of the robot body. The new capabilities will improve the ability of robots to operate effectively and safely in unconstrained environments and also their ability to communicate and co-operate with each other and with humans
Defying the Rules: How Self-Organising Systems Work
The goal of the project is to identify general rules and conditions for emergent self-regulation by studying multiple diverse instances of self-regulating social systems in humans, ants and robots. The theories developed will be applied to develop improved methodologies for solving related problems such as community regeneration and automated manufacturing.
Forming Abstract Concepts through Hierarchical History Compression
This project studies ways of compressing a robot’s history of observations and actions in a hierarchy of temporal sequences. The goal of the project is to scale up instance-based reinforcement learning algorithms by allowing a large amount of historical data to be efficiently stored and utilised for hidden state identification and future reward estimation. The project is inspired by recent theories of how information is encoded in the human cortex. Our research hypothesis is that this form of compression will produce a number of abstract concepts on different timescales and that it will be possible to relate these concepts to those used by animals and humans.
Learning to Reach with Crude Corrective Movements
This project has modified a reinforcement algorithm previously published by Andrew Fagg to allow a simulated iCub robot to learn reaching in three dimensions with a four DOF arm. Initial results show very fast learning from a fixed starting point to a fixed goal point. Future work will use this algorithm with further learning algorithms to learn a general learning capability from an arbitrary arm configuration to an arbitrary goal point.
Cyber Savannah Competition for Webots
This project will develop a competition framework for the Webots development environment, produced by Cyberbotics Ltd. It is based upon a savannah styled environment, with robot inhabitants that represent herbivorous animals. The goal of the competition is to accumulate the largest store of energy within the tribe. This is achieved through efficient feeding behaviour and gaining more robot members for the tribe.