This research presents a spatio-temporal graph framework for recognizing complex human activities. It enhances robotic response capabilities by accurately interpreting interaction patterns from continuous video input in collaborative and care-assistive environments.

Takeshi Yujiro Sakamoto, Isabelle Joy Lambert, Priyansh Rajan Tripathi, Zheng Hua Lin, Dylan Marcus Price

This paper introduces a zero-shot visual grounding model for autonomous agents. It utilizes CLIP embeddings and context-aware modules to identify target objects in open environments without prior training, supporting versatile mission objectives in unstructured terrains.

Zihan Meiling Fang, Reuben Isaac Carver, Ashwin Karthik Menon, Olivia Grace Shelton, Tao Ming Xun

This study presents a deep network architecture for visual SLAM. Structural correspondences are learned to identify revisited locations in visually cluttered scenes, boosting accuracy and robustness of loop closure detection for mobile autonomous platforms.

Ritika Sanjana Radhakrishnan, Ezra Julian Brooks, Mei Hua Shen, Carlos Nathaniel Rhodes, Nobuyuki Daichi Okamoto

This paper proposes a day-night image translation model. Cycle-consistent transformers align multispectral domains, allowing autonomous systems to maintain visual continuity across lighting conditions for persistent surveillance and object tracking applications.

Yu Chenlong Zhao, Edward Julian Vance, Ramesh Sharvan Iyer, Emily Renee Baxter, Shohei Takeshi Yamada

This paper introduces a compression method for 3D point clouds. Sparse attention filters reduce data size, while viewpoint-invariant sampling ensures quality preservation for restoration during autonomous spatial mapping and terrain analysis tasks.

Karthik Narayan Venkatesh, Fiona Mae Harrison, Liang Yuwei Zhang, Hiroto Masashi Takeda, Alicia Grace Whitmore

This paper introduces a tracking algorithm designed for occluded, crowded scenes. It uses scene-flow constraints and temporal attention reweighting to preserve trajectory accuracy across frames in high-density autonomous interaction spaces.

Neelabh Ajay Trivedi, Aurora Faith Hayes, Xu Ling Zhi, Marco Felix Donovan, Keisuke Ryoji Nakamoto

This study presents a depth completion method for sparse sensor input. Scale-adaptive pyramids improve geometric consistency, allowing autonomous vehicles to generate dense depth maps in challenging outdoor environments with minimal input data.

Anish Tejasvi Nambiar, Chloe Isabelle Sanders, Zhu Li Ping, Masato Hiroshi Kaneda, Natalie Rose McAllister