Email
Password
Remember meForgot password?
    Log in with Twitter

article image'Ambidextrous' robots can accelerate speed e-commerce

By Tim Sandle     Jan 19, 2019 in Technology
A test study has shown how so-termed 'ambidextrous robots' can significantly speed e-commerce. By applying artificial intelligence, the robot is capable of grasping a diverse range of object shapes without training.
As e-commerce continues to expand, retailers are seeking more efficient ways to service consumers with goods and service as efficiently and rapidly as possible. There are also projects underfoot by companies like Amazon, Walmart and Alibaba to find alternatives to using human labor for next-generation warehouses, designed to expand the volume and variety of goods.
Robots are a solution to the demand for automation. However, while robotics has advanced in many areas there is one aspect where robotics still faces a challenge. This the ability of a machine to reliably grasp a diverse range of products. Most robots used in e-commerce use suction grippers, which have limitations when it comes to manipulating certain types of objects. Overcoming this challenge is the focus of engineers working at University of California - Berkeley.
Commenting on the problem, researcher Jeff Mahler explains: "Any single gripper cannot handle all objects. For example, a suction cup cannot create a seal on porous objects such as clothing and parallel-jaw grippers may not be able to reach both sides of some tools and toys."
The UC Berkeley engineers have adopted an “ambidextrous” approach which aims to be compatible with a range of gripper types. The approach also introduces a framework based on a common “reward function” for each gripper type. This sets out to quantify the probability that each gripper will succeed. This enables the robot to make a fast decision as to which gripper to use for any given situation.
To compute the reward function for any gripper, the researchers have developed a process for learning reward functions which involves training on large synthetic datasets. The datasets are quickly generated using structured domain randomization together with analytic models of sensors and engineering design that takes into account the physics and geometry of each gripper.
A development model of the robot - a parallel-jaw gripper and a suction cup gripper fitted to a two-armed robot - was able to remove all objects from a bin, containing 25 objects previously unseen by the robot, at a rate of over 300 picks per hour. The robot achieved a 95 percent reliability rating.
The research has been published in the journal Robotics, with the research paper headed "Learning ambidextrous robot grasping policies."
More about Robots, Robotics, Ecommerce, Sales
More news from
Latest News
Top News