Soft material picking control by robot arm

Exhibition event

  • ROBODEX Nagoya (October 2021)


  • Robot arm hand gripping strength is adjusted by reinforcement learning to transport soft objects (bread) without damaging their shape.
  • Assuming automation in a bakery, we have developed a reinforcement learning AI that grips soft objects (bread) with “just the right amount of force” and moves and arranges them without damaging their shape.


  1. reate a pan model using AGX Dynamics for physics representation, and place the pan model on Unreal Engine4.
  2. Reinforcement learning was used to infer the grasping force to hold the bread without crushing it.
  3. After grasping the bread, programmatically instruct the bread model to move to another location.
  4. Repeat steps 2 and 3 to align the bread models by bread type.


  • AGX Dynamics
  • Unreal Engine 4

Back to list