Humanoid Robot Control Using Spiking Neural Networks

In recent years there has been an increasing amount of interest in spiking neural networks, which are more biologically realistic than rate based models. Whilst a lot of simulation work has been carried out on spiking neural networks, relatively little research has been done on the use of spiking neural networks to control robots in an environment. The main obstacle to the development of embodied spiking neural networks is the lack of an easy way to link spiking neural network simulators to a robot body in an environment. To address this issue we have developed iSpike: an open source software interface between a humanoid robot, the iCub, and a spiking neuron simulator NeMo. iSpike performs encoding of sensory stimulus received from the iCub into patterns of neuron spikes used by the simulator. The spike output from the simulator is decoded into motor commands for the robot to perform. In our implementation we attempt to mimic the visual, proprioceptive and motor control pathways of the human body. The implementation has been tested with the iCub virtual simulator, the test demonstrates the possibility of real-time control and monitoring of the virtual iCub robot using NeMo. Performance tests show that motor control and proprioceptive pathway performance is efficient enough for real-time application. iSpike is currently available at http://ispike.sf.net.