Semester of Graduation
Summer 2026
Degree Type
Dissertation
Degree Name
Interdisciplinary Engineering
Committee Chair/First Advisor
Yan Fang
Second Advisor
Jian Zhang
Third Advisor
Beibei Jiang
Fourth Advisor
Bobin Deng
Abstract
This dissertation presents a neuromorphic, event-driven approach to visual simultaneous localization and mapping (SLAM) for energy-efficient operation on edge devices. Conventional SLAM relies on frame-based processing and computationally intensive algorithms that limit deployment on power-constrained platforms such as embedded robots and microscale autonomous systems. To address this, the work develops a biologically inspired framework combining event-based vision sensing with neuromorphic computation to reduce redundant processing while maintaining robust localization and mapping. The system integrates an event-driven visual odometry front-end with a topological mapping back-end, connected through a compact scene representation based on vector symbolic architectures. It was implemented and benchmarked on the NVIDIA Jetson Orin Nano and the Raspberry Pi 5, where per-module latency and energy consumption were measured to characterize deployability. A targeted PyTorch acceleration of the dense linear-algebra stages further improved latency and energy efficiency on the Jetson Orin Nano, and the framework was validated on real-world event-camera datasets for localization accuracy, robustness, and energy efficiency. Building on this single-agent foundation, the dissertation extends the framework to collaborative multi-agent neuromorphic SLAM through a Multi-Agent Error Correction (MAEC) approach. By enabling agents to share compact environmental information and cooperatively refine their spatial estimates, the proposed method improves localization consistency and mapping performance while maintaining low communication overhead. This collaborative capability preserves the energy-efficient characteristics of the underlying neuromorphic architecture and demonstrates the potential of scalable multi-agent SLAM for resource-constrained edge robotic systems.