The Neurorobotics Platform connects spiking neural networks to virtual and real robots. This enables embodiment experiments in both neuroscience and AI.
The Neurorobotics Platform (NRP) was initially developed for virtualized brain research, offering a complete toolchain with which neuroscientists can test and refine their models by designing experiments and then carrying them out in simulation. For example, with the NRP, one could model place cells in the hippocampus and then study a virtual animal’s navigation or sensorimotor skills as it operates in a completely simulated environment (e.g., a maze or a straight or sinusoidal vertical path). While these closed-loop simulations linking perception, cognition and action are tremendously potent tools in a neuroscience lab, their usefulness extends far beyond.
Indeed, the capabilities of the NRP (simulation of a physically realistic world, design and simulation of robotic models) enable it to perform virtual prototyping of robots as well as transfer learning, i.e. development of robotic controllers through learning in simulation followed by implementation into a physical robot. These robots can then be readily built as real machines and function like the simulated ones. This approach will not only speed up robot development by orders of magnitude, it will also dramatically improve the testing and verification of robotic behavior within a wide variety of circumstances. This will be particularly important in scenarios where robots physically operate in close proximity to, or even interact with humans (increased safety), or in scenarios where physical damage to the robot may stem from incorrect task performance (lower experimental cost).
Finally, the NRP has the potential to become a potent tool in the field of artificial intelligence (AI), especially insofar as the concept of embodied cognition becomes prominent. This concept postulates that intelligence, artificial or otherwise, is shaped by aspects of the body (sensors and actuators) to which it has access to. By providing the software architecture to exploit the potential of AI in robotics while ensuring no risk comes with it, the NRP makes a decisive contribution to embodied AI, and paves the way for novel robotic control technologies that achieve robustness and adaptation capabilities far beyond current algorithmic controls… and ones that actually rival biologic systems. While the actual implementation details of these technologies remain to be decided, the NRP is perfectly positioned to contribute to standard digital techniques (e.g. deep learning) as well as to low-power alternatives such as neuromorphic computing.