Scepter914 Website

For Japanese

Biography

Profile

  • Name: Satoshi Tanaka

Work Experience

  • Aug. 2025 - Freelance engineer
  • Apr. 2020 - Jul. 2025, TIER IV, Inc.
    • Jun. 2024 - Jul. 2025 Lead Software Engineer for MLOps in perception module
    • Apr. 2020 - Dec. 2023 Software Engineer for Autonomous Driving
  • Internship
    • Apr. 2018 - Apr. 2019, Internship at Preferred Networks, Inc. as a part-time engineer
    • Aug. 2017 - Mar. 2018, Internship at Hitachi, Ltd as a research assistant

Academic Background

  • Master’s Degree in Information Science and Engineering, the University of Tokyo
    • Apr. 2018 - Mar. 2020, Ishikawa Senoo Lab, Department of Creative Informatics, Graduate School of Information Science and Technology
  • Bachelor’s Degree in Precision Engineering, the University of Tokyo
    • Apr. 2017 - Mar. 2018, Kotani Lab, Research Center for Advanced Sceience and Technology
    • Apr. 2016 - Mar. 2018, Dept. of Precison Engineering
    • Apr. 2014 - Mar. 2016, Faculty of Liberal Arts

Interest

  • Robotics, Computer Vision, Control Theory
  • Building autonomous robotic systems that can interact with the physical world faster and more dexterously than humans
    • Real-time 3D object detection
    • Development of mechanisms capable of fast and compliant motion
    • Force control for dynamic manipulation

Publication

International Conference (First author)

  • Satoshi Tanaka, Keisuke Koyama, Taku Senoo, Makoto Shimojo, and Masatoshi Ishikawa, High-speed Hitting Grasping with Magripper, a Highly Backdrivable Gripper using Magnetic Gear and Plastic Deformation Control, 2020 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS2020), Proceedings, pp. 9137 - 9143. [2020 IEEE Robotics and Automation Society Japan Joint Chapter Young Award]
  • Satoshi Tanaka, Keisuke Koyama, Taku Senoo, and Masatoshi Ishikawa, Adaptive Visual Shock Absorber with Visual-based Maxwell Model Using Magnetic Gear, The 2020 International Conference on Robotics and Automation (ICRA2020), Proceedings, pp. 6163-6168.
  • Satoshi Tanaka, Taku Senoo, and Masatoshi Ishikawa, Non-Stop Handover of Parcel to Airborne UAV Based on High-Speed Visual Object Tracking, 2019 19th International Conference on Advanced Robotics (ICAR2019), Proceedings, pp. 414-419.
  • Satoshi Tanaka, Taku Senoo, and Masatoshi Ishikawa, High-speed UAV Delivery System with Non-Stop Parcel Handover Using High-speed Visual Control, 2019 IEEE Intelligent Transportation Systems Conference (ITSC19), Proceedings, pp. 4449-4455.

ArXiv papers (First author)

  • Satoshi Tanaka, Koji Minoda, Fumiya Watanabe, Takamasa Horibe, Rethink 3D Object Detection from Physical World, arXiv 2025, https://arxiv.org/abs/2507.00190.
  • Satoshi Tanaka, Samrat Thapa, Kok Seang Tan, Amadeusz Szymko, Lobos Kenzo, Koji Minoda, Shintaro Tomie, Kotaro Uetake, Guolong Zhang, Isamu Yamashita, Takamasa Horibe, AWML: An Open-Source ML-based Robotics Perception Framework to Deploy for ROS-based Autonomous Driving Software, arXiv 2025, https://arxiv.org/abs/2506.00645.

International Conference (Not first author)

  • Taisei Fujimoto, Satoshi Tanaka, and Shinpei Kato, LaneFusion: 3D Object Detection with Rasterized Lane Map, the 2022 33rd IEEE Intelligent Vehicles Symposium (IV 2022), Proceedings, pp. 396-403.

Other publication

  • Kazunari Kawabata, Manato Hirabayashi, David Wong, Satoshi Tanaka, Akihito Ohsato AD perception and applications using automotive HDR cameras, the 4th Autoware workshop at the 2022 33rd IEEE Intelligent Vehicles Symposium (IV 2022)

Award, Scholarship

Projects

(Research) Rethink 3D Object Detection from Physical World

AWML

DepthAnything-ROS



(Research) High-speed Hitting Grasping with Magripper

  • Developed high-speed hitting grasping executed seamlessly from reaching with Magripper, a highly backdrivable gripper, and hitting grasping, high-speed grasping framework.
  • Accepted at IROS2020 [2020 IEEE Robotics and Automation Society Japan Joint Chapter Young Award]

(Research) Adaptive Visual Shock Absorber with Magslider

  • Developed visual shock absorber system with high-speed vision, high-backdrivablilty hardware, and force control.
  • Accepted at ICRA2020

(Research) High-speed supply station for UAV delivery system

  • Developed high-speed supply station for UAV delivery system
  • Accepted at ITSC2019


Robotic Competition

  • Team Leader for ABU Robocon2016
  • Winner of National Championships, 2nd-runnerup of ABU Robocon, ABU Robocon award.
  • Visited to the prime minister’s residence as the team leader of representation from Japan team. Reported by link and link.

Other projects

Latest change (blog, survey)

OW-Adapter: Human-Assisted Open-World Object Detection with a Few Examples (IEEE Transaction on visualization and computer graphics 2024/01)
OW-Adapter: Human-Assisted Open-World Object Detection with a Few Examples (IEEE Transaction on visualization and computer graphics 2024/01) Summary https://www.youtube.com/watch?v=QNub6PYMp1k Method 非常にイケているUI 所謂 pseudo label の 分布の表現も必要っぽい Experiment Discussion
2024/08/11 · Surveys
SAM 2: Segment Anything in Images and Videos (arxiv 2024/07)
SAM 2: Segment Anything in Images and Videos (arxiv 2024/07) Summary https://github.com/facebookresearch/segment-anything-2 参考 https://speakerdeck.com/tenten0727/segment-anything-model-2?slide=5 Method dataset Experiment 高速化 FPSが結構出てるのすごい Model Size (M) Speed (FPS) SA-V test (J&F) MOSE val (J&F) LVOS v2 (J&F) sam2_hiera_tiny 38.9 47.2 75.0 70.9 75.3 sam2_hiera_small 46 43.3 (53.0 compiled*) 74.9 71.5 76.4 sam2_hiera_base_plus 80.8 34.8 (43.8 compiled*) 74.7 72.8 75.8
2024/08/11 · Surveys
FRNet: Frustum-Range Networks for Scalable LiDAR Segmentation (arxiv2023/12)
FRNet: Frustum-Range Networks for Scalable LiDAR Segmentation (arxiv2023/12) Summary https://github.com/Xiangxu-0103/FRNet https://xiangxu-0103.github.io/FRNet Frustum basedな 3D semantic segmentation Method KNN post-processing の差 Experiment 可視化 https://www.youtube.com/watch?v=PvmnaMKnZrc https://www.youtube.com/watch?v=4m5sG-XsYgw https://www.youtube.com/watch?v=-aM_NaZLP8M incorrectの量が減っている score Semi-supervised: 少ないデータセットでもちゃんと上手く
2024/08/10 · Surveys
SIMPL: A Simple and Efficient Multi-agent Motion Prediction Baseline for Autonomous Driving (arxiv2024/02, RA-L 2024)
SIMPL: A Simple and Efficient Multi-agent Motion Prediction Baseline for Autonomous Driving (arxiv2024/02, RA-L 2024) Summary https://www.youtube.com/watch?v=_8-6ccopZMM https://github.com/HKUST-Aerial-Robotics/SIMPL?tab=readme-ov-file Transformer base prediction Method Experiment 3060Ti 250Actorでも20ms程度はめっちゃ使い勝手は良さそう 可視化の動画めっちゃ良さげに見え
2024/08/04 · Surveys
A Survey on Autonomous Driving Datasets: Data Statistic, Annotation, and Outlook (arxiv2024/01)
A Survey on Autonomous Driving Datasets: Data Statistic, Annotation, and Outlook (arxiv2024/01) Summary https://github.com/daniel-bogdoll/ad-datasets 元になったrepository Method 結構偏りが激しいし、結局nuScenesが良さげに見える task一覧としてわかり
2024/08/02 · Surveys
DepthAnything-ROS
Summary Developed by hobby coding. Repository: https://github.com/scepter914/DepthAnything-ROS Made prototype ROS2 package of DepthAnything with TensorRT C++. DepthAnything is one of the high performance monocular depth estimation. This work is covered by official github repository and its gallary. Performance results Model Params RTX2070 TensorRT Depth-Anything-Small 24.8M 27 ms, VRAM 300MB Depth-Anything-Base 97.
2024/05/06 · Projects
DepthAnything-ROS
Summary 趣味開発 Repository: https://github.com/scepter914/DepthAnything-ROS TensorRT C++ を用いて、DepthAnything の ROS2 パッケージのプロトタイプを作成。 DepthAnything は高性能な単眼深度推定手法の1つ 本プロジェクトは
2024/05/06 · Projects
DepthAnythingのROS2 packageを作った
DepthAnythingのROS2 packageを作った 概要 DepthAnything のROS2 packageを作った それに対する感想や周辺Toolを作った話をつ
2024/02/08 · Blogs
VSCode Vim で日本語の変換が変になる問題への対処
VSCode Vim で日本語の変換が変になる問題への対処 概要 VSCode Vim で日本語の変換が変になる問題への対処 環境 Ubuntu 22.04 VSCode v1.82.0 VSCode Vimのextensionを入れている状態
2023/09/11 · Blogs
MatrixVT: Efficient Multi-Camera to BEV Transformation for 3D Perception (arxiv2022/11)
MatrixVT: Efficient Multi-Camera to BEV Transformation for 3D Perception (arxiv2022/11) Summary https://github.com/Megvii-BaseDetection/BEVDepth BEVDepthの後継、軽量version 軽量なBEV-base Camera 3d detection CPUでも動作するレベルで軽量 CPUでも数10
2022/12/20 · Surveys