Spatial Representation and Path Planning for Autonomous Robots: A Convolutional Neural Network Approach

Zhixin Zhao

Simen Owen Academic Proceedings Series, 2026, vol. 7, 130-140

Abstract: Accurate environment modeling and efficient path planning are fundamentally critical for autonomous robots navigating complex and highly dynamic spatial layouts, such as automated warehouses and industrial manufacturing floors. However, traditional search-based algorithms and standard reinforcement learning models frequently fail to efficiently process high-dimensional spatial data. This deficiency inevitably leads to substantial computational overhead in rapidly changing dynamic environments or results in kinematically infeasible paths that are characterized by dangerous wall-hugging and erratic steering behaviors. To comprehensively address these persistent limitations, this study proposes a novel joint framework integrating a lightweight multi-scale Convolutional Neural Network (CNN) with a Deep Q-Network (DQN). Specifically, the CNN architecture systematically extracts hierarchical spatial features from two-dimensional local grid maps, effectively compressing the state space required for the DQN without losing critical geometric information. Additionally, a sophisticated composite reward function, which strictly enforces safety clearances and ensures kinematic smoothness, is introduced to optimally guide the planning policy. Evaluated rigorously across a diverse set of dynamic simulation scenarios (n=100), the proposed method achieved an impressive 93.8% task success rate alongside a remarkably low collision rate of 4.1%. Furthermore, the system maintained a real-time planning latency of 25 ± 3 ms and successfully reduced the Maximum Angular Velocity Variance to 0.62 rad/s2, thereby significantly improving kinematic feasibility when compared to existing baseline models. Ultimately, these empirical findings demonstrate that the proposed multi-scale spatial representation effectively balances perception accuracy with computational efficiency, providing a highly robust and readily deployable navigation framework suitable for advanced edge-computing robotic systems.

Keywords: path planning; neural networks; reinforcement learning; environment modeling; autonomous navigation (search for similar items in EconPapers)
Date: 2026
References: Add references at CitEc
Citations:

Downloads: (external link)
https://soapubs.com/index.php/SOAPS/article/view/2276/2090 (application/pdf)

Related works:
This item may be available elsewhere in EconPapers: Search for items with the same title.

Export reference: BibTeX RIS (EndNote, ProCite, RefMan) HTML/Text

Persistent link: https://EconPapers.repec.org/RePEc:axf:soapsa:v:7:y:2026:i::p:130-140

Access Statistics for this article

More articles in Simen Owen Academic Proceedings Series from Scientific Open Access Publishing
Bibliographic data for series maintained by Yuchi Liu ().