Skip to content

Latest commit

 

History

History
54 lines (36 loc) · 3.07 KB

keypoint_visualization.md

File metadata and controls

54 lines (36 loc) · 3.07 KB

Introduction

This file mainly illustrates how the affordance keypoints are used in the real-world physical experiments. As mentioned in the manuscript, there are five in total keypoints for each affordance. The keypoints 1 and 2 determine the operational direction of the object part. The keypoints 3 and 4 are auxiliary points, which help complete the geometry of the object part. They are important especially for the grasp affordance. As shown in the Fig. 1, the keypoints 3 and 4 help determine the rotation angle of the robot arm to achieve top-down grasp. The keypoint 5 is the center point of the object part.

Grasp

As shown in the Fig. 1, the operational location is computed by the keypoint 5 with depth information. For all physical expeirments, top-down grasp is employed. The operation direction, which is acutally the rotation angle of the robot wrist, is determined by the keypoint 3 and 4.

image

Fig. 1 - Operational location and direction converted from affordance keypoints for the grasp affordance.

Contain

For the contain affordance, there is no operational direction. The operational location is computed by the keypoint 5 with depth information.

image

Fig. 2 - Operational location and direction converted from affordance keypoints for the contain affordance.

Cut

For the cut affordance, it requires determining where and how to cut. The operational location is computed by the keypoint 5 with depth information. The operational direction is computed by the keypoints 1 and 2.

image

Fig. 3 - Operational location and direction converted from affordance keypoints for the cut affordance.

Pound

For the pound affordance, it requires determining where and how to pound. The operational location is computed by the keypoint 5 with depth information. The operational direction is computed by the keypoints 1 and 2.

image

Fig. 4 - Operational location and direction converted from affordance keypoints for the pound affordance.

Scoop

For the scoop affordance, it requires determining where and how to cut. The operational location is computed by the keypoint 5 with depth information. The operational direction can be determined by either affordance mask or keypoints. By the affordance mask, the operational direction can be determined by the relative position between the grasp and scoop affordance. By the affordance keypoint, the operational direction is computed by the keypoints 1 and 2.

image

Fig. 5 - Operational location and direction converted from affordance keypoints for the scoop affordance.

Wgrasp

Similar to the grasp affordance, we only employ 2.5D pose for wrap-grasp. The wrap-grasp pose is always straightly facing at the target object. The operational location is computed by the keypoint 5 with depth information.

image

Fig. 6 - Operational location and direction converted from affordance keypoints for the wgrasp affordance.