This research introduces a tracking framework for aerial systems. Temporal zoom windows isolate relevant motion, while compact encoding modules predict pedestrian trajectories, reducing compute overhead during long-duration drone surveillance operations.

Aditya Nalin Sharma, Grace Eliza McConnell, Kaiwei Shen Fu, Junpei Hiro Tanabe, Samantha Louise Griffin

This paper presents a scene completion model for autonomous assistance systems. Missing image regions are restored using partial inpainting, then projected spatially to reconstruct full 3D scenes for mobility support and safe route estimation.

Nilesh Rajan Chandra, Fiona Mae Roberts, Tao Liang Sheng, Aurelien Jean Pascal, Hailey Juliet Smith

This research presents an unsupervised keypoint detection model. Using cross-view image matching and category-based learning, robots can infer object semantics and manipulate tools or items more accurately across previously unseen orientations.

Raghavendra Kishore Dutt, Eleanor Lily Sanderson, Guo Rui Xin, Takeshi Hideo Mori, Beatrice Claire Lewis

This study presents a vision-based terrain assessment system. Edge-aware disparity maps and surface segmentation enable AGVs to evaluate ground traversability and adapt motion paths in outdoor, semi-structured environments under varying conditions and sensor noise levels.

Pranav Karthik Shenoy, Louisa Anne Gregory, Wen Qiang Zhao, Atsushi Kenji Sakamoto, Heidi Rachel Langford

This paper introduces a continual learning framework for autonomous agents. Object-centric memory and multi-stage alignment modules allow robots to update visual knowledge interactively, improving adaptability during incremental, real-world object exploration tasks.

Ritesh Manohar Deshmukh, Harriet Lucy Bradford, Zhao Ling Wen, Makoto Hiroshi Yamada, Alice Claire Rosenthal

This work presents a scene flow model combining multi-stream optical flow with depth fusion. Spatiotemporal priors enhance prediction consistency, enabling motion-aware navigation in crowded or complex driving environments for autonomous platforms.

Sachin Narayan Pillai, Julia Violet Cameron, Zhen Yu Liu, Hiroshi Daisuke Tanaka, Camille Elise Ferguson

This study proposes a segmentation framework for indoor robotics. Top-down refinement and contextual pooling modules allow precise object separation in cluttered settings, facilitating robotic navigation, interaction, and autonomous grasp planning in dense domestic environments.

Nirav Sudhir Patil, Zoe Harper Dalton, Wang Tian Cheng, Kazuki Masaru Fujimoto, Elise Marion McKinney