Last updated November 8, 2019 at 1:57 pm
A new project is training robots to think on the move, and it might be the key to making them grasp items more effectively.
Why This Matters: A lot of robots are rubbish at everything except one specific task. We need to broaden their abilities.
When we play games, like Jenga or Pick Up Sticks, we don’t sit still. Nor do we close our eyes, think and then blindly grasp at the objects in a futile attempt to win a game. Instead, we move around, craning our heads to try and find the easiest target.
That’s what a new project from the Australia Centre for Robotic Vision is trying to teach robots when grasping for objects.
“Our aim at the Centre is to create truly useful robots able to see and understand like humans. So, in this project, instead of a robot looking and thinking about how best to grasp objects from clutter while at a standstill, we decided to help it move and think at the same time,” says PHD researcher Doug Morrison.
A world-first approach to grasping
The project’s ‘active perception’ approach is the first in the world to focus on real-time grasping by stepping away from a static camera position or fixed data collecting routines.
Instead, the project builds a ‘map’ of grasps in a pile of objects, which continually updates as the robot moves. The same way we see and consider new ways to tackle the Jenga stack as we see things from different perspectives, this real-time mapping predicts the quality and pose of grasps at every pixel in a depth image.
“Like humans, this allows the robot to change its mind on the go in order to select the best object to grasp and remove from a messy pile of others.”
“The beauty of our active perception approach is that it’s smarter and at least ten times faster than static, single viewpoint grasp detection methods,” says Morrison.
Morrison has tested and validated his active perception approach at the Centre’s QUT-based Lab in ‘tidy-up’ trials using a robotic arm to remove 20 objects, one at a time, from a pile of clutter. His robots could successfully complete the task 80 per cent of the time, 12 per cent better than other robots using a traditional single viewpoint.
What allows the robot to approach the task more like a human is the Multi-View Picking (MVP) controller. As the robot reaches to grasp an object, the MVP continuously analyses the objects. This allows the robot to see more high-quality grasps than a normal fixed camera.
“By looking at a pile of objects from multiple viewpoints on the move, a robot is able to reduce uncertainty caused by clutter and occlusions,” says Morrison.
“It also feeds into safety and efficiency by enabling a robot to know what it can and can’t grasp effectively. This is important in the real world, particularly if items are breakable, like glass or china tableware messily stacked in a washing-up tray with other household items.”
It’s not just about grasping an object; the robot needs to do something with it
The next step for these grasping robots moves away from safe and effective grasping and into the realm of meaningful vision-guided manipulation. The researchers want the robot to not only grasp an object but to then do something with it.
“Take, for example, setting a table, stacking a dishwasher or safely placing items on a shelf without them rolling or falling off,” says Morrison.
However, Morrison also wants to fast-track how quickly a robot learns to grasp the physical object and wants to challenge it with a set of weird shapes.
“It’s funny because some of the objects we’re looking to develop in simulation could better belong in a futuristic science fiction movie or alien world – and definitely not anything humans would use on planet Earth!”
Some might ask: what’s the point? However, there is method in this scientific madness.
“At first glance, a stack of human household items might look like a diverse data set, but most are pretty much the same. For example, cups, jugs, flashlights and many other objects all have handles, which are grasped in the same way and do not demonstrate difference or diversity in a data set,” explains Morrison.
“We’re exploring how to put evolutionary algorithms to work to create new, weird, diverse and different shapes that can be tested in simulation and also 3D printed.
“A robot won’t get smarter by learning to grasp similar shapes. A crazy, out-of-this-world data set of shapes will enable robots to quickly and efficiently grasp anything they encounter in the real world.”