• 2004 – A. Miller, P. Allen – “GraspIt: A Versatile Simulator for Robotic Grasping”
• 2007 – M. Ciocarlie, C. Goldfeder, P. Allen – “Dimensionality reduction for hand-independent dexterous robotic grasping”
• 2008 – Ciocarlie, Allen – “On-Line Interactive Dexterous Grasping”

Minimizing a Distance function

• Sum of distances between desired contact points and object surface
• Sum of angular differences between surface normals and hand contact normals
• Successfully creates a grasp, however grasping is not certain

Simulated annealing

• Takes into account the potential of hand posture
• Calculate wrenches at each desired grasp point(Grasp Wrench Space)
• Take the convex hull of the GWS
• Quality is determined by the distance from the origin to the closest boundary of the GWS
• If the origin is not contained in the space, no force closure (Q=0).

Eigengrasps

• http://www.robotics.cs.rpi.edu/nems/2007/slides/ciocarlie.pdf - good presentation on method
• Eigengrasps provide the space across which to search, but still too big
• 173 seconds for 10000 iterations
• How can we reduce that time and provide some interactivity with people at the same time?
• Human + Robot!!

Grasp Sequence

1. Human positions wrist, reducing the space of eigengrasps drastically.
2. Automated grasp planning algorithm (simulated annealing) finds a stable pregrasp along the wrist’s direction of motion to reduce the search space.
3. Contact-detection algorithm in GraspIt? ! is used to concurrently calculate the final grasp position.
4. User indicates acceptance of grasp by either moving wrist to new position or accepting grasp.
5. Final acceptance comes with a close command, which executes the final grasp position

Links

• Columbia University Robotics Lab - GraspIt!
• Related video