Neurorobotic Models of Spatial Navigation and Their Applications
Spatial navigation requires many parallel cognitive processes, from low-level perception to high-level planning and decision-making. Findings from neuroscience provide inspiration for spatial navigation in robotic applications, improving the efficiency and adaptability of the systems. For example, neuromorphic hardware mimics the computational strategy of the brain for event-driven, massively parallel processing of information. In this dissertation, a working demonstration of a ground robot performing road following in an outdoor environment is introduced, with control computations run on neuromorphic hardware. Next, a neuromorphic path planning algorithm that accounts for environmental costs is presented, using spiking wave propagation inspired by hippocampal function. On top of perception and path planning, a memory system is necessary for contextual awareness and adaptation when navigating in a dynamic environment. To address this, a model of the hippocampus, medial prefrontal cortex, and neuromodulatory areas is discussed, in which the model is able to rapidly learn new information when consistent with a familiar contextual schema and prevent catastrophic forgetting of previously learned tasks. The model is successfully demonstrated on the Toyota Human Support Robot, tasked with finding and retrieving objects in multiple contexts. In summary, these models and demonstrations show how neurobiological systems of navigation can be used to improve robotic navigation at many levels of processing, which in turn reveals more insights on how biological systems navigate.