Research Pillars — Tohoku Robotic Systems Laboratory

Our research builds on a shared vision of working robot intelligence — robots that are useful working systems capable of perceiving, manipulating, moving, cooperating with humans, and adapting to changing environments. We study robots that handle real-world complexity: difficult-to-model objects, flexible materials, uncertainty, and unstructured terrain.

The laboratory organizes its work around four research pillars: manipulation planning and learning for deformable objects, motion learning and planning for mobile manipulators, action support by autonomous robots, and connected robotics with edge AI. Each pillar contributes to the broader goal of building intelligent robotic systems that support human work and society.

Manipulation Planning and Learning of Deformable Objects

Tasks involving deformable objects can be observed in various environments such as living spaces, factory settings, and logistics sites, but it is not easy to have automated machines perform these tasks. Our laboratory aims to research and systematize the modeling, recognition, manipulation, and behavioral learning of flexible objects. We are tackling this using various methods, including traditional image processing techniques, motion planning extensions, deep learning, reinforcement learning, and imitation learning.

Representative topics

generation of cloth manipulation procedures based on predictions
automatic acquisition of folding tasks
simulations for deformable objects
differentiable simulation

Motion Learning and Planning for Mobile Manipulators

This research focuses on the intelligence of robots equipped with robotic arms mounted on mobile platforms, known as mobile manipulators. Mobile manipulators have both mobility and object manipulation capabilities, allowing for various applications. However, they need to address the redundancy of movement degrees of freedom and the complexity of the surrounding environment. Our laboratory is advancing several studies from the perspectives of vision, motion planning, and behavioral learning.

Representative topics

system integration of household support robots
efficiency of object recognition actions based on long-term activity experience
simultaneous execution of task planning and motion planning

Action Support by Autonomous Robots

This research focuses on robot systems that support human tasks by staying close to people. The requirement for robots here is not to proceed at their own pace, but to recognize and predict human actions and intentions, and to move appropriately in accordance with the person. Our laboratory is working on supporting daily activities such as dressing and fetching objects.

Representative topics

dressing assistance system for individuals with hemiplegia
proposal and implementation of branch-type robots for task support
object manipulation skills by imitating human actions

Connected Robotics and Edge AI

Connected Robotics extends intelligent robotic systems from individual robots to networks of robots, IoT sensors, tools, and edge AI modules. This research direction investigates how distributed physical agents can share observations, construct semantic world models, reason under uncertainty, and generate explainable actions in real-world environments. Rather than treating sensors and edge devices as passive data sources, this pillar studies how they can become active reasoning nodes that support robot decision-making, adaptive sensing, and human-understandable explanations.

Representative topics

robot-IoT collaboration
distributed sensing and adaptive sensor placement
shared semantic world models
semantic evidence sharing and belief updating
edge AI and small language models for robot reasoning
explainable collective decision-making

Projects

Explore student projects

See how students contribute across all four research pillars.

Student Projects →