Using Geometry to Detect Grasp Poses in 3D Point Clouds ten Pas, - - PowerPoint PPT Presentation

▶
using geometry to detect grasp poses in 3d point clouds
SMART_READER_LITE
LIVE PREVIEW

Using Geometry to Detect Grasp Poses in 3D Point Clouds ten Pas, - - PowerPoint PPT Presentation

Aug 30, 2023 286 likes •654 views

Using Geometry to Detect Grasp Poses in 3D Point Clouds ten Pas, Platt Northeastern University September 15, 2015 Helping Hands Lab Objective Three possibilities: Instance-level grasping Category-level grasping Novel object

slide-1
SLIDE 1

Using Geometry to Detect Grasp Poses in 3D Point Clouds

ten Pas, Platt Northeastern University September 15, 2015

Helping Hands Lab

slide-2
SLIDE 2

Objective

Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

slide-3
SLIDE 3

Objective

Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

The robot has a detailed description of the object to be grasped.

slide-4
SLIDE 4

Objective

Grasp the banana Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

The robot has general information about the object to be grasped.

slide-5
SLIDE 5

Objective

Grasp the thing in the box

The robot has no information about the object to be grasped.

Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

slide-6
SLIDE 6

Objective

“Easier” “Harder” Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

slide-7
SLIDE 7

Objective

Most research assumes this “Easier” “Harder” Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

slide-8
SLIDE 8

Objective

Our focus:

  • 1. Grasping novel or partially

known objects

  • 2. Robustness in clutter

Three possibilities: – Instance-level grasping – Category-level grasping – Novel object grasping

Related Work:

  • 1. Fischinger and Vincze. Empty the basket - a shape based learning approach for grasping piles of unknown
  • bjects. IROS'12.
  • 2. Fischinger et al. Learning grasps for unknown objects in cluttered scenes. IROS 2013.
  • 3. Jiang et al. Efficient grasping from rgbd images: Learning using a new rectangle representation. IROS 2011.
  • 4. Klingbeil et al. Grasping with application to an autonomous checkout robot. IROS 2011.
  • 5. Lenz et al. Deep learning for detecting robotic grasps. RSS 2013.
slide-9
SLIDE 9

Differences to Prior Work

– Localizing 6-DOF poses instead of 3-dof grasps – Point clouds obtained from multiple range sensors instead of a single RGBD image – Systematic evaluation in clutter

slide-10
SLIDE 10

Novel Object Grasping

slide-11
SLIDE 11

Novel Object Grasping

Input: a point cloud Output: hand poses where a grasp is feasible.

slide-12
SLIDE 12

Novel Object Grasping

Input: a point cloud Output: hand poses where a grasp is feasible.

Each blue line represents a full 6- DOF hand pose

slide-13
SLIDE 13

Novel Object Grasping

Input: a point cloud Output: hand poses where a grasp is feasible.

– don't use any information about object identity Each blue line represents a full 6- DOF hand pose

slide-14
SLIDE 14

what was there what the robot saw

Why Novel Object Grasping is Hard

slide-15
SLIDE 15

Why Novel Object Grasping is Hard

slide-16
SLIDE 16

what was there what the robot saw (monocular depth) what the robot saw (stereo depth)

Why Novel Object Grasping is Hard

slide-17
SLIDE 17
  • 2. Classification

Our Algorithm has Three Steps

  • 3. Outlier removal
  • 1. Hypothesis generation
slide-18
SLIDE 18
  • 2. Classification

Our Algorithm has Three Steps

  • 3. Outlier removal
  • 1. Hypothesis generation
slide-19
SLIDE 19

Step 2: Grasp Classification

We want to check each hypothesis to see if it is an antipodal grasp

slide-20
SLIDE 20

… then we could check geometric sufficient conditions for a grasp

If we had a “perfect” point cloud...

We would check whether an antipodal grasp would be formed when the fingers close

slide-21
SLIDE 21

If we had a “perfect” point cloud...

But, this is closer to reality...

Missing these points!

So, how do we check for a grasp now?

slide-22
SLIDE 22

If we had a “perfect” point cloud...

But, this is closer to reality...

Missing these points!

So, how do we check for a grasp now?

Machine Learning (i.e. classification)

slide-23
SLIDE 23

Classification

We need two things:

  • 1. Learning algorithm + feature representation
  • 2. Training data
slide-24
SLIDE 24

Classification

We need two things:

  • 1. Learning algorithm + feature representation

– SVM + HOG – CNN

  • 2. Training data
slide-25
SLIDE 25

Classification

We need two things:

  • 1. Learning algorithm + feature representation

– SVM + HOG – CNN

  • 2. Training data

– automatically extract training data from arbitrary point clouds containing graspable objects

slide-26
SLIDE 26

Training Set

slide-27
SLIDE 27

Training Set

97.8% accuracy (10-fold cross validation)

slide-28
SLIDE 28

Test Set

slide-29
SLIDE 29

Test Set

94.3% accuracy on novel objects

slide-30
SLIDE 30

Experiment: Grasping Objects in Isolation

slide-31
SLIDE 31

Results: Objects Presented in Isolation

slide-32
SLIDE 32

Experiment: grasping objects in clutter

slide-33
SLIDE 33

Results: Clutter

73% average grasp success rate in 10-object dense clutter

slide-34
SLIDE 34

Conclusions

  • New approach to novel object grasping
  • Use grasp geometry to label hypotheses

automatically

  • Average grasp success rates:
  • 88% for single objects
  • 73% in dense clutter
slide-35
SLIDE 35

Questions?

atp@ccs.neu.edu http://www.ccs.neu.edu/home/atp

ROS packages – Grasp pose detection: wiki.ros.org/agile_grasp – Grasp selection: github.com/atenpas/grasp_selection