top of page

Research Tuesday

  • Writer: Yicheng Z.
    Yicheng Z.
  • 11 minutes ago
  • 2 min read

07-15-2025


As I planned, Tuesday is about the research. In this kind of post, I would like to talk about the current study I'm doing. This is also a process that helps me list out some of the thought process in the research. Maybe by putting them out here, I can clean up my mind a little bit.


Please bear with me with my format of citation. I won't reach anything too specific.

Robot in DOM tasks

My current study topic is about the deformable object manipulation. This area of robotics is one of the peripheral area of robotics. It's not something as basic as control, nor is something as high level as reinforcement learning or path planning algorithms. DOM tasks are tasks that robots interact with deformable objects. While it seems specific and niche, it can be quite general since virtually everything in this world is 'somewhat' deformable. Rigidity is just a simplified state of the actual reality.

The capability to robustly interact with deformable objects is important for the autonomy of robots. However, current development of this is still a work in progress. There are four ways that are usually used to tackle this DOM problem.

  • Specifically designed end-effector. (for specific scenario, the gripper is customized and optimized for the task)

  • Physics model based control (a physics model is built and is used to anticipate the deformation of the object.

  • Data-driven methods (usually end-to-end method, that use reinforcement learning to figure the mapping between current action, current sensor input and next step.

  • Hybrid: A hybrid of previous several.

Each of those methods have their unique drawbacks

  • Task-specific end-effector design: trade off generality.

  • Physics model based model: computation hungry, and leads to high latency.

  • Data-driven method: the data need to be generated in simulation workspace, since it would not be feasible to have a deformable object set up in real life to let the robot explore. Thus, the virtual-reality gap became a problem.


New solution is needed, and we seek to get inspiration from human.

We will continue on Thursday!


Picture of the day

Flower at home, got from a clearance area in the Lowe's. It's somehow withering these days.
Flower at home, got from a clearance area in the Lowe's. It's somehow withering these days.

 
 
 

Comments


bottom of page