Scalable Imitation Learning
Learning robust manipulation policies from large-scale demonstrations and representation-rich observations.
Robot Learning • Physical AI • Foundation Models
Makoto Sato 佐藤 誠人
I am a second-year Ph.D. student in the Department of Technology Management for Innovation, Graduate School of Engineering, The University of Tokyo. My research focuses on robot learning, with an emphasis on scalable imitation learning and test-time scaling for robot foundation models.
Building robot learning systems that scale across data, simulation,
and inference-time reasoning.
About
Researching more capable and general-purpose robot behavior
My work sits at the intersection of imitation learning, reinforcement learning, task and motion planning, and robot foundation models. I am especially interested in how large-scale data, simulation, and inference-time computation can improve long-horizon reasoning and contact-rich manipulation.
At the University of Tokyo, I study scalable imitation learning and test-time scaling for robot foundation models. Before starting my Ph.D., I worked on task and motion planning for contact-rich manipulation at Nara Institute of Science and Technology, and on multimodal object identification at Saitama University.
My research experience spans academia, industry, and applied AI settings, including Sakana AI, AIRoA, Matsuo Institute, and AIST. Across these environments, I have focused on embodied intelligence, simulation-driven learning, and scalable training pipelines for robotics.
Research Interests
- Robot learning
- Imitation learning
- Task and motion planning
- Reinforcement learning
- World models
- Robot foundation models
- Contact-rich manipulation
- Vision-language models
Toolbox
- Python
- C++
- PyTorch
- ROS
- Docker
- MuJoCo
- Isaac Sim
- PyBullet
- SAPIEN
- Gazebo
- CARLA
Research
Current directions
I am interested in building robot learning systems that scale with richer demonstrations, stronger simulation, and more effective inference-time decision making.
Test-Time Scaling for Robot Models
Studying how additional compute at inference time can improve in-context decision making for embodied agents.
Long-Horizon Manipulation
Combining task and motion planning with learning-based policies for precise and contact-rich manipulation tasks.
Simulation and World Models
Using generative simulation, world models, and embodied datasets to train more general robot policies.
Education
Academic path
My background combines mechanical engineering, information science, and technology management, with a consistent focus on robot learning.
The University of Tokyo
Ph.D. Student, Department of Technology Management for Innovation, Graduate School of Engineering
Research: Scalable Imitation Learning and Test-Time Scaling for Robot Foundation Models
Nara Institute of Science and Technology
M.Eng., Graduate School of Advanced Science and Technology, Information Science
Research: Task and Motion Planning for Contact-Rich Manipulation
Saitama University
B.Eng., Faculty of Engineering, Department of Mechanical Engineering and System Design
Research: Object Identification Using Visual and Surface-Tracing Vibration Signals
Experience
Research, teaching, and industry work
My experience includes research collaborations, seminar teaching, and engineering work in both academic and industrial environments.
Research Affiliations
- Sakana AI, Inc. 2025 - Present
- AI Robot Association (AIRoA) 2025 - 2026
- Matsuo Institute, Inc. 2020 - 2025
- AIST 2020
Teaching
- Physical AI Spring Seminar Kinematics and AI models
- Deep RL Spring Seminar Model-based reinforcement learning and world models
- World Model Course Simulation and computer graphics
Industry
- Sony Semiconductor Solutions Corporation Full-time engineer, 2024
I value projects that connect reliable engineering with ambitious research problems in embodied intelligence.
Publications
Selected papers and drafts
A selection of recent work in imitation learning, world models, and task and motion planning. See the CV for a fuller list.
SAIL: Test-Time Scaling for In-Context Imitation Learning with VLM
Makoto Sato, Yusuke Iwasawa, Yujin Tang, So Kuroki
arXiv draft, 2026
Imitation Learning with Mid-Level Representations for Object Rearrangement
Makoto Sato and collaborators
JSAI, 2022
Scaling Laws of Model Size for World Models
Makoto Sato and collaborators
JSAI, 2023
Scaling Laws of Dataset Size for VideoGPT
Masahiro Negishi, Makoto Sato, and collaborators
JSAI, 2023
Construction and Validation of Action-Conditioned VideoGPT
Koudai Tabata, Junnosuke Kamohara, Makoto Sato, and collaborators
JSAI, 2023
Task and Motion Planning Using Residual Reinforcement Learning for Long-Horizon Precise Object Manipulation Task
Makoto Sato, Yuhwan Kwon, Yoshihisa Tsurumine, Takamitsu Matsubara
SCI, 2024
Projects
Selected ongoing and recent projects
Project themes range from uncertainty-aware planning to simulation-led data generation for large robot models.
Exploratory Task and Motion Planning Using Uncertainty-Aware VLM
Studying how uncertainty-aware vision-language models can generate exploratory subgoals for task and motion planning.
Large-Scale Demonstration Generation for VLA via Generative Simulation
Building data generation pipelines for training autonomous robot foundation models with generative simulation.
Large-Scale Imitation Learning for Bimanual Dexterous Manipulation
Exploring scalable imitation learning for humanoid and dexterous manipulation in simulation-rich environments.
Long-Horizon Manipulation with Task and Motion Planning
Investigating planning-driven robot manipulation for long-horizon tasks that require precise interaction and sequencing.
Contact
Get in touch
I am happy to discuss research collaborations, internships, and projects related to robot learning, physical AI, and scalable embodied intelligence.
Email
makoto.sato@weblab.t.u-tokyo.ac.jp
GitHub
github.com/makolon
Curriculum Vitae
Download PDF
Location
Tokyo, Japan