Home

Sloman, A.
From "Baby Stuff" to the World of Adult Science: Developmental AI from a Kantian viewpoint

---

Howard A. M. & Bekey G. A.
Intelligent Learning for Deformable Object Manipulation
Autonomous Robots, Kluwer Acadademic Publ, 2000, 9, 51-58

---

Uses:

• Bags filled with sponge, cotton, sand and water
• Mass-spring like physical model [Kelvin Model]
• Self calibration
• Stereo visual 3D input & feedback
• Neural network with backpropagation for generalization of the learned behaviour

The idea is actually simple but powerfull:

Two microbots (arms) are used to press against both sides of the object, their movement is determined by setting the possitions to which they must be located, (this possitions are provided by the vision system). The force the microbots detect (as they press against the object's surfaces) and the object's deformation are measured and used to estimate the required coeficients for the material's model.

Various attempts to lift the object are performed, starting with a force equal to the object weight, after every failure the force to be used is increassed, until the cameras register a succesfull lifting. The results are saved in a table.

Note that the choice of the starting force and the increment for every attempt where conveniently decided by the programmer.

A neural network receives as inputs the estimated parameters for every material:

• the deformability coefficient,
• damping coefficient and
• object weight

It will be trained and used to predict the expected lifting force for new material coefficients.

Saha, M.; Isto, P. & Latombe, J. C.
Motion planning for robotic manipulation of deformable linear objects
Experimental Robotics, 2008, 39, 23-32

Saha, M. & Isto, P.
Motion planning for robotic manipulation of deformable linear objects
2006 Ieee International Conference On Robotics And Automation (Icra), Vols 1-10, 2006, 2478-248

Lou, Y. H. & Nelson, B. J.
Fusing Force and Vision Feedback for Manipulating Deformable Objects
Journal of Robotic Systems, 2001, 18, 103-117

---

Uses:

• Finite Element Analysis (commercial software ANSYS)
• Snakes (to represent the contour of the object)
• Requires models of the camera (pinhole, the distance to the object must be known in advance) and force sensors

They worked with rubber Buna-N., assuming they are elastic and nearly incompressible (which leads to some reasonable numerical errors).

FEM works on stable states.

ANSYS is used to determine the model constants from training examples.

In summary:

They use the force sensors to estimate FEM parameters during the training phase. The acquired material model is used to parameterize the snakes for contour tracking.

From the contour information, the material model is used to estimate forces, therefore obtaining a force sensor from observed visual deformations.

Cremers, D.
Dynamical statistical shape priors for level set-based tracking
IEEE TRANSACTIONS ON PATTERN ANALYSIS AND MACHINE INTELLIGENCE, IEEE COMPUTER SOC, 2006, 28, 1262-1273

---

Morris, D. & Salisbury, K.
Automatic preparation, calibration, and simulation of deformable objects [old version]
Computer Methods In Biomechanics And Biomedical Engineering, 2008, 11, 263-279

---

Uses:

• Meshes
• Bayesian approach: particle filters

Burion, S.; Conti, F.; Petrovskaya, A.; Baur, C. & Khatib, O.
Identifying physical properties of deformable objects by using particle filters
2008 Ieee International Conference On Robotics And Automation, Vols 1-9, 2008, 1112-1117

---

Uses:

• Mass-spring skeleton
• Bayesian approach: particle filters

Brief and nice background section, specially for the beginer in the area. Covers FEM, mass-spring systems, particle filters and other references.

The physics of the objects whose properties they want to identify are simulated. They do not give details about the simulation software.

Observe that they perform measurements at steady states (where things are easier). Instead of comparing distances between models, they look at resulting forces at each mass node (they should be zero for a stateady state).