Hugo de Garis - Utah-Brain Project

The aim of the Utah-Brain project is to build an artificial brain which will control the behaviour of a life-sized robotic cat (image right). The brain will contain around 75 million artificial neurons which live in a specially built, highly parallel computer. The kitten will be able to see, hear, and feel. Its behaviours, such as walking and playing with a ball etc, will be remotely controlled by the artificial brain via a radio link. All of these behaviours will be evolved, rather than pre-programmed by a human operator.

The project is being conducted at the Utah State University, USA and led by Professor Hugo de Garis.

The Artificial Brain:

The brain of the robotic cat is a computer called the CAM Brain Machine (CBM). The core of the CBM consists of 72 micro-processors. These processors are of a type known as FPGA's. FPGA's (Field Programmable Gate Arrays) are computer processors which don't have fixed logic circuits. Instead, the logic can be configured by the user for a wide variety of applications. In the CBM, the FPGA's have been configured to implement a 3-dimensional neural network. The CBM can implement a neural net module of 1152 neurons. Roughly 64,000 of these modules are then time-shared over the array of processors. This highly parallel architecture allows the CBM to process artificial neural networks extremely quickly. In fact, the CBM is estimated to be as fast at implementing the neural net algorithm as 10,000 Pentium II 400MHz processors.

The CBM operates in one of two modes: evolution mode, and run mode. It is during evolution mode that the behaviours of the cat will be learnt. The neural networks are grown and evolved using a genetic algorithm. The CBM is equiped with hardware modules dedicated to performing this genetic algorithm. The evolved network configurations are then saved in the CBM's 1 Gigabyte of RAM. During run mode the evolved neural nets process input stimuli from the cat and respond with the appropriate behaviours.

Current Status:

December 1999 saw the completion and delivery of the brain hardware (the CBM). Throughout 2000/2001 experiments were performed to understand the machine's capabilites in terms of evolving neural nets. These met with encouraging results. The main tasks for 2002 will be to design and evolve the 64 thousand or so modules which will make up the artificial brain. It is hoped that the intelligent robotic cat will be completed within a few years.

Future Plans

The Utah-Brain team is certainly ambitious in its plans for the future. Despite the fact that the CBM/robokitten concept has yet to be proven, plans are already afoot for the building of a second generation "brain machine". The aim is to build a machine capable of implementing a billion artificial neurons by around 2005/2006.

Links:

Prof. Hugo de Garis
Genobyte (CBM Manufacturer)
Senapps