Robot Learning Manipulation Action Plans by “Watching” Unconstrained Videos from the World Wide Web

From Yezhou Yang, Yi Li, Cornelia Fermuller and Yiannis Aloimonos:

In order to advance action generation and creation in robots beyond simple learned schemas we need computational tools that allow us to automatically interpret and represent human actions. This paper presents a system that learns manipulation action plans by processing unconstrained videos from the World Wide Web. Its goal is to robustly generate the sequence of atomic actions of seen longer actions in video in order to acquire knowledge for robots. The lower level of the system consists of two convolutional neural network (CNN) based recognition modules, one for classifying the hand grasp type and the other for object recognition. The higher level is a probabilistic manipulation action grammar based parsing module that aims at generating visual sentences for robot manipulation.


The list of the grasping types.

Experiments conducted on a publicly available unconstrained video dataset show that the system is able to learn manipulation actions by “watching” unconstrained videos with high accuracy.... (article at Kurzweilai.net) (original paper)

Comments (0)

This post does not have any comments. Be the first to leave a comment below.


Post A Comment

You must be logged in before you can post a comment. Login now.

Featured Product

maxon motor’s - Exoskeleton Joint Actuator

maxon motor's - Exoskeleton Joint Actuator

A brushless DC motor solution for use in hip and knee exoskeletons. This complete joint actuation unit consists of motor, gearhead, encoder and position controller. Fitting absolute encoder directly at the joint rotation provides designers increased positioning accuracy. The unit will deliver 54Nm of continuous torque and 120Nm on a 20% duty cycle. The system can be operated on supplies between 10 and 50V DC and the actuation speed is up to 22rpm.