Using rat navigation models to learn orientation from visual input on a mobile robot

Rodents possess extraordinary navigation abilities that are far in excess of what current state- of- the- art robot agents are capable of. This paper describes research that is part of larger project aimed at developing a robot navigation system that is capable of robust autonomous navigation in real- time by using biologically plausible constructs inspired from the many neurological and behavioral studies conducted on freely navigating rats. Specifically, this paper discusses the implementation of a ratinspired system that allows a robot to learn and recognize its allocentric orientation based on what it perceives. The system described here is a pragmatic, minimalist implementation of recently proposed models of the rat head direction system. The rat head direction system are collectively, the neural assemblies that neuroscientific studies suggest are responsible for encoding the orientation of the rat's head in a global reference frame. This paper describes an implementation of the system on an autonomous robot that operates in real time, in office environments, and with laboratory limited computational hardware. Experiments conducted in a messy laboratory environment with a range of visual features demonstrate that the demonstrate that the system, while simple in structure, is able to learn and recognize allocentric orientations after 5 minutes of exploration. The results demonstrate the worth of bridging robotics with biological research and pave the way for developing a more complete, and competent, robot architecture.