Improving the Performance of an Insect-Inspired Navigation Model Using Directional Selective Collision Detection

Autonomous navigation is an essential capability for both robots and animals, enabling them to achieve their locomotion objectives within specific environments. Drawing inspirations from nature, there has been recent introduction of an insect-inspired algorithm for autonomous navigation. This model integrates global working memory inspired by sweat bee path integration (PI) and local cues derived from the locust giant motion detector (LGMD). Experimental findings provide evidence of the effectiveness of this integrated approach in addressing navigation tasks, involving both stationary and moving obstacles. Furthermore, the algorithm demonstrates efficient computation, independent of external data or environment dependency. However, there is scope for improvement concerning collision avoidance and autonomous navigation performance. To address these concerns, two solutions are proposed in this study: 1) employing the Directional Selective Neuron (DSN) model as an alternative to LGMD for local cues, offering collision signals and visual object motion direction to enhance collision avoidance motion control, and 2) introducing a fusion mechanism that integrates DSN output with PI global memory to enhance navigation performance. Simulation data reveals the following outcomes: 1) in static environments, complete avoidance of collisions during foraging-homing and a significant decrease in navigation task completion time, and 2) in dynamic environments, notably improved collision avoidance robustness and navigation efficiency. These results effectively demonstrate the enhancement achieved through the adoption of the proposed bio-inspired method and exemplify the principle of learning from nature.