Manipulation in Clutter

Part 2:                                                   

Grasp Selection       

MIT 6.421

Robotic Manipulation

Fall 2023, Lecture 9

Follow live at https://slides.com/d/FZziNhk/live

(or later at https://slides.com/russtedrake/fall23-lec09)

Course Roadmap

1. Basic pick and place:
   • single known object
   • assuming known pose
2. Geometric perception
   • single known object
   • unknown pose
3. This week
   • ​​many/diverse unknown objects
   • unknown poses
   • simple task: "clutter clearing"

"Clutter clearing" (an early project at TRI)

Initial conditions using simulation

Contact mechanics (friction cones, hydroelastic)

Grasp analysis / optimization

The red figure is the Minkowski sum of blue and green figures.

https://en.wikipedia.org/wiki/Minkowski_addition

$$A \oplus B = \{a+b \,|\, a \in A, b \in B\}$$

(unpublished) work by Lucas Manuelli ~2017.  No deep learning -- this is geometry only.

Voxelization

http://www.open3d.org/docs/latest/tutorial/Advanced/voxelization.html

Limitations of using geometry only

• No understanding of what an object is.
  • "Double picks"
  • Might pick up a heavy object from one corner
• Partial views
• Depth returns don't work for transparent objects
• ...