PhD Dissertation Defense by Dhanuja Tharith WANNIARACHCHIGE | Enhancing Multi-view, Multi-modal Sensing, Perception and Actuation For Edge Intelligence

Enhancing Multi-view, Multi-modal Sensing, Perception and Actuation For Edge Intelligence

ABOUT THE TALK

Artificial Intelligence of Things (AIoT) technologies have revolutionized intelligent sensing within Cyber-Physical Systems (CPS), powering applications such as large-scale surveillance and autonomous transportation. These systems rely heavily on vision-based Deep Neural Networks (DNNs), yet executing such complex models on resource-constrained edge devices remains a challenge. Transmitting high-bandwidth sensor data to the cloud is impractical, while lightweight DNNs deployed locally often compromise accuracy. This dissertation addresses this core tradeoff by introducing the concept of collaborative DNN inference, where multiple IoT nodes with overlapping Fields-of-View (FoVs) share intermediate representations or compact features to enhance perception quality while minimizing bandwidth usage.

The research spans three major contributions. ComAI introduces feature-level collaboration among stationary cameras, enabling small edge DNNs to achieve detection accuracy close to that of larger models while sustaining high throughput. SteerCam extends this idea to active sensing, combining deep fusion with Reinforcement Learning (RL)–based camera steering to dynamically adjust FoVs and create intelligent overlaps that maximize both individual and collective accuracy. Finally, FusionBridge generalizes collaborative inference to multimodal (RGB + LiDAR) and mobile settings, leveraging the MultiSense-RL-Arena simulation platform. It employs cross-modal feature fusion, where 3D LiDAR embeddings refine 2D visual detections for improved geometric reasoning.

Together, these efforts demonstrate that collaborative deep fusion can significantly elevate perception accuracy, robustness, and efficiency in distributed edge AI systems—establishing a new paradigm for intelligent, cooperative sensing in future AIoT deployments

SPEAKER BIOGRAPHY

Dhanuja Wanniarachchi is a Ph.D. Candidate in Computer Science, under the supervision of Professor Archan MISRA. His research focuses on enhancing Edge Artificial Intelligence in Networked Settings. His PhD research addresses the challenges in multi-view, multimodal sensing, perception, and actuation for enhancing edge intelligence, utilizing lower complexity models to achieve higher perception in networked edge-based applications. His work was accepted for publication at IEEE INFOCOM 2022 and ACM UbiComp / ISWC 2025. During his candidature, he did an internship at A*STAR, Singapore. He previously earned his bachelor’s degree from the University of Moratuwa, majoring in Electronics and Telecommunications.