Ms Alexandria Wampamba, Dr Mansour Hakim-Elahi,
Volume 15, Issue 4 (12-2025)
Abstract
The deployment of autonomous robots in unstructured, cluttered environments remains a significant challenge, particularly for low-cost platforms. While the Dynamic Window Approach (DWA) provides a robust foundation for reactive navigation, its performance is often suboptimal due to a lack of historical context, leading to oscillatory behavior and entrapment in local minima. This paper presents a novel, cost-effective mechatronic system that enhances DWA with a real-time spatial memory module and optimizes its performance using a Bayesian Optimization strategy. Our platform integrates a Raspberry Pi 4 with a fused ultrasonic and infrared sensor suite. The core innovation is a Local Occupancy History Map that provides a short-term, decaying memory of obstacle locations. This memory influences the DWA’s trajectory evaluation, discouraging paths through recently occupied space. Furthermore, we employ Bayesian Optimization loop to automatically tune the critical hyperparameters of the navigation system—the memory decay rate and the history weight—to maximize efficiency and safety. We validate our system in complex indoor environments, comparing the baseline DWA, the DWA with Spatial Memory (DWA-SM), and the optimized DWA-SM (DWA-SM-Opt). Quantitative results demonstrate that the optimized system (DWA-SM-Opt) achieves a 40% reduction in average path completion time and a 65% decrease in collisions compared to the baseline DWA. Qualitative analysis confirms more intelligent, fluid navigation and a consistent ability to escape trapping configurations. This work establishes that the fusion of a lightweight spatial memory with an AI-driven optimization routine, implemented on low-cost hardware, can yield a level of performance previously associated with more complex and expensive systems.
