SpikingNav: Robust Embodied Navigation with Spiking Neural Policies

SpikingNav is a spiking-neural framework for indoor embodied navigation, combining a Spiking Sensing Encoder (SSE) with a recurrent Spiking Policy Network (SPN). The model is evaluated on PointNav and ObjectNav under clean observations and visual corruptions. According to the paper summary, it maintains competitive clean performance while using fewer parameters and less per-step computation than a matched ANN baseline. ObjectNav success improves from 31.05% to 34.12%, while average success under visual corruptions rises from 8.45% to 13.71%. The authors also validate the spike-based sensing method on the Thruster-V2 neuromorphic chip.