Tsubasa Hirakawa
Papers
13
Total Citations
150
H-Index
6
About
Tsubasa Hirakawa is a computer vision and robotics researcher whose work sits at the intersection of deep learning, visual explanation, and intelligent manipulation systems. His research primarily advances object detection, semantic segmentation, and robotic grasping, with a particular focus on making neural network decision-making interpretable and practically deployable. Hirakawa is perhaps best known for the MT-DSSD framework, a multi-task deconvolutional single shot detector that simultaneously performs object detection, semantic segmentation, and grasp-point estimation within a unified network — a significant contribution to logistics robotics that has accumulated 50 citations across its iterations. Equally influential is his Attention Branch Network, which bridges visual explanation and performance improvement by leveraging attention mechanisms to both interpret and enhance CNN predictions, garnering 34 citations since 2018. His commitment to explainability extends into deep reinforcement learning, where he has developed attention-based methods for visualizing agent decision-making in actor-critic architectures, addressing the critical "black-box" problem in autonomous systems. Additional contributions span natural language generation for domestic service robots and automated data collection pipelines that reduce costly manual labeling. Collectively, Hirakawa's body of work reflects a sustained effort to build robotic systems that are not only capable, but transparent and practically scalable.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4
- 5
- 6
- 7
- 8Iterative Coarse-to-Fine 6D-Pose Estimation Using Back-propagation5 citations · 2021
- 9
- 10