Evaluation of Object Affordance Detection

We empirically evaluate our grasp methodology using a real robot. We designed a grasp scenario that the robot first picks-up an object, and then carries the object to the placing position to see if the object slips due to bad grasp or not. A particular grasp was considered a success if the robot is able to perform this scenario successfully. In this experiment, 40 household objects were used which are depicted at the beginning of the video.
We first demonstrated how to grasp the first six objects (i.e.,
number 1 to 6) with the same affordance label as illustrated at the beginning of the demonstration. Then, the robot tried to grasp each object four times. In the first round, we used the learned grasp templates to grasp all objects and calculate the success rate. In the
second round, we taught two additional grasp templates to the robot including objects no. 7 and no. 8 and compute the success rate to see the improvement in grasping over time.