Iori Yanokura
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Iori Yanokura has a total of 29 co-authors. Here are the top co-authors by number of shared publications:
- Kei Okada
The University of Tokyo
3 shared publications - Kento Kawaharazuka
The University of Tokyo
2 shared publications Masayuki Inaba
The University of Tokyo
2 shared publications- Masayuki Inaba
Department of Mechano-Informatics, The University of Tokyo, Tokyo, Japan
2 shared publications - Bogdan Franczyk
University of Leipzig
1 shared publication - Haruki Kozuka
The University of Tokyo
1 shared publication
Latest Publications
- Open Access
Grapes Ready for Picking: Investigating the Readiness of Precision Viticulture Datasets for Scene Reconstruction
Public datasets of vineyard imagery were collected for detection, segmentation and disease mapping and are now proposed as a source for 3D scene reconstruction. Their documentation rarely states the properties that reconstruction depends on, such as view overlap, parallax and frame spacing, so it is unclear which can be reconstructed. This study measures that for the publicly available vineyard datasets. A structured search identifies eighteen datasets. Ten are characterized under a schema of thirty-six documented and measured properties, and only six contain overlapping views. Every sequence that registers is reconstructed with one fixed pipeline of structure from motion and 3D Gaussian splatting, giving twenty-seven reconstructions on two held-out splits. Registration is decided by the capture rather than by the dataset; novel-view quality varies as much within a dataset as between datasets; and frame spacing is the one capture property that limits reconstruction. Three preprocessing steps are tested as ablations on the failure cases. Band alignment restores registration, whereas dehazing a backlit pass and masking the chassis of a ground robot give no improvement that this study can resolve. The schema and pipeline apply to any vineyard dataset, and the results are published as an open online catalog.
Open AccessGrapes Ready for Picking: Investigating the Readiness of Precision Viticulture Datasets for Scene Reconstruction
Public datasets of vineyard imagery were collected for detection, segmentation and disease mapping and are now proposed as a source for 3D scene reconstruction. Their documentation rarely states the properties that reconstruction depends on, such as view overlap, parallax and frame spacing, so it is unclear which can be reconstructed. This study measures that for the publicly available vineyard datasets. A structured search identifies eighteen datasets. Ten are characterized under a schema of thirty-six documented and measured properties, and only six contain overlapping views. Every sequence that registers is reconstructed with one fixed pipeline of structure from motion and 3D Gaussian splatting, giving twenty-seven reconstructions on two held-out splits. Registration is decided by the capture rather than by the dataset; novel-view quality varies as much within a dataset as between datasets; and frame spacing is the one capture property that limits reconstruction. Three preprocessing steps are tested as ablations on the failure cases. Band alignment restores registration, whereas dehazing a backlit pass and masking the chassis of a ground robot give no improvement that this study can resolve. The schema and pipeline apply to any vineyard dataset, and the results are published as an open online catalog.
- 23 May 2025
- Published by Taylor & Francis in Advanced Robotics
- Vol. 39 (11) , 647-662
GenAerialNav: obstacle avoidance in real flight for generalized multirotors by reinforcement learning with variable Acc-Properties in dynamics
- 23 May 2025
- Published by Taylor & Francis in Advanced Robotics
- Vol. 39 (11) , 647-662
GenAerialNav: obstacle avoidance in real flight for generalized multirotors by reinforcement learning with variable Acc-Properties in dynamics
- Open Access
- 3 April 2025
- Published by Taylor & Francis in Advanced Robotics
- Vol. 39 (7) , 323-337
Situation classification of living environment by daily life support robot using pre-trained large-scale vision-language model
- 3 April 2025
- Published by Taylor & Francis in Advanced Robotics
- Vol. 39 (7) , 323-337
Open AccessSituation classification of living environment by daily life support robot using pre-trained large-scale vision-language model
- Open Access
Understanding Action Sequences based on Video Captioning for Learning-from-Observation
Learning actions from human demonstration video is promising for intelligent robotic systems. Extracting the exact section and re-observing the extracted video section in detail is important for imitating complex skills because human motions give valuable hints for robots. However, the general video understanding methods focus more on the understanding of the full frame,lacking consideration on extracting accurate sections and aligning them with the human's intent. We propose a Learning-from-Observation framework that splits and understands a video of a human demonstration with verbal instructions to extract accurate action sequences. The splitting is done based on local minimum points of the hand velocity, which align human daily-life actions with object-centered face contact transitions required for generating robot motion. Then, we extract a motion description on the split videos using video captioning techniques that are trained from our new daily-life action video dataset. Finally, we match the motion descriptions with the verbal instructions to understand the correct human intent and ignore the unintended actions inside the video. We evaluate the validity of hand velocity-based video splitting and demonstrate that it is effective. The experimental results on our new video captioning dataset focusing on daily-life human actions demonstrate the effectiveness of the proposed method. The source code, trained models, and the dataset will be made available.
Open AccessUnderstanding Action Sequences based on Video Captioning for Learning-from-Observation
Learning actions from human demonstration video is promising for intelligent robotic systems. Extracting the exact section and re-observing the extracted video section in detail is important for imitating complex skills because human motions give valuable hints for robots. However, the general video understanding methods focus more on the understanding of the full frame,lacking consideration on extracting accurate sections and aligning them with the human's intent. We propose a Learning-from-Observation framework that splits and understands a video of a human demonstration with verbal instructions to extract accurate action sequences. The splitting is done based on local minimum points of the hand velocity, which align human daily-life actions with object-centered face contact transitions required for generating robot motion. Then, we extract a motion description on the split videos using video captioning techniques that are trained from our new daily-life action video dataset. Finally, we match the motion descriptions with the verbal instructions to understand the correct human intent and ignore the unintended actions inside the video. We evaluate the validity of hand velocity-based video splitting and demonstrate that it is effective. The experimental results on our new video captioning dataset focusing on daily-life human actions demonstrate the effectiveness of the proposed method. The source code, trained models, and the dataset will be made available.
- 31 December 2018
- Published by Springer Nature
- p. 56-66
Variance Based Trajectory Segmentation in Object-Centric Coordinates
- 31 December 2018
- Published by Springer Nature
- p. 56-66
Variance Based Trajectory Segmentation in Object-Centric Coordinates