Efficient Multi-Sensor Neural Docking for Resource-Constrained Autonomous Systems: An Empirical Study on Aerial Platforms
This paper explores a lightweight, camera-free approach to autonomous docking for resource-constrained aerial systems using four directional infrared (IR) proximity sensors and a compact neural network. The proposed approach formulates docking as a 15-class classification task, mapping four sensor readings to discrete docking actions. The model consists of a single hidden layer with 16 neurons and is designed for embedded deployment through INT8 quantization and TensorFlow Lite Micro. The study describes a 5,000-sample dataset, neural network architecture, quantization pipeline, and proposed microcontroller-oriented hardware design. The reported classification accuracy reaches 100% on the experimental dataset. The dataset appears to be synthetic or rule-derived, and real-world sensor-noise validation, closed-loop flight testing, and actual docking success-rate evaluation remain future work. Therefore, the reported accuracy should not be interpreted as evidence of real-world autonomous docking performance. This work presents a preliminary systems-oriented exploration of TinyML, sensor-based navigation, and resource-efficient machine learning for autonomous robotic platforms.
Authors
- Himanshu Patel
- Rajnish Singh
Institutions
- Gautam Buddha University (IN)
- Manipal University Jaipur (IN)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-13
- DOI
- https://doi.org/10.5281/zenodo.24183693
- Primary Topic
- Advanced Neural Network Applications
- Type
- article
- Field-Weighted Citation Impact
- 0.00