Cobots are Coming

How many of you know what a robot looked like 50 years ago? Well this is shaky the robot developed at Stanford and we just celebrated Shakey’s fifty fifth birthday. Now if I ask you to think about a robot fifty years into the future, what pops into your head? Maybe an industrial robot, but if we’re talking about robots that are at work every day right now, then we’re probably talking about something more like this (factory robots), pretty impressive, right? but I know what you’re thinking when you see impressive factory robots, deep down inside you’re secretly wondering, where’s my robot? I want a robot to cook for me, to clean for me to do my laundry, I mean these robots are building a Tesla why can’t they do my dishes?

The answer is really simple, and don’t take it too personally, but you are the problem.

Humans are just too dynamic in nature for traditional robots to be able to keep up. While we can have robots do incredibly complex tasks, like building entire planes, these robots are only following very specific pre-programmed instructions on how to stamp out metal, where to rivet, and how to weld. Now lets look at a kitchen, you’ve got people running around, you’ve got a different layout for every kitchen, and you have people working alongside the robot, messing everything up. Almost makes building the plane sound easier, right?

Luckily, this will not be the case for long, as Cobots, a new class of robots designed specifically to work alongside people, have reached the age of implantation

“collaborative robots”, known by their shorthand “cobots” are meant to operate in conjunction with, and in close proximity, to humans to perform their tasks. Unlike their more isolated counterparts, cobots are intentionally built to physically interact with humans in a shared workspace

What is a cobot

We can see how they are meant to operate in close proximity to humans in the very way they are trained. Rather than programmed to a specific set of steps using programming tools, many cobots are trained by humans manipulating the arms and training by example. Humans control the bot by physically moving it around, with the cobot remembering the steps and perhaps even the end goal of what is being accomplished, and then repeats those steps, optimizing them to achieve increasingly better outcomes.

But that’s only the beginning, last time Shay was up on this stage, he briefly told you about flippy, the hamburger making robot, so today, I’m going to tell you about what really makes flippy special. While the machinery inside of it is incredibly complex, Flippy’s greatness comes from it’s ai that allows it to work alongside humans.

Dubbed Miso AI, Flippy is able to see, serve, and move unlike any robot before. So, what does MisoAI do that’s so new? It has intuition. It contains an intuitive element to ensure that the operations of a cobot can be composed or altered dynamically by an operator, or simply put, it allows a human to control a robot far more easily.

Flippy has 6 axes of motion, something that is very hard to be able to control with just one joystick. Sure we can use inverse kinematics and a host of other complex mathematics to program it’s movements by computer, but for dynamic environments (like a burger joint) that are constantly changing it’s incredibly resource intensive to do so. So, misoAI’s machine learning algorithms undergo a process known as reducing dimensionality.

The process of dimensionality reduction begins with an operator moving the robot arm through various task-specific motions, like flipping a burger, essentially training it how to move in a more fluid and useful way in a given context.

All of the data that is generated by these movements is then fed through a neural network that first compresses the data into two dimensions, and then decodes that compressed representation to try to recreate the initial. It’s a little bit like the Generative Adversarial Networks that we studied earlier in the ai module, where the ai is constantly trying to match the movements of the pre-fed expert data

And that’s it, all that needs to happen next is give the robot two-dimensional instructions on a joystick, and the robot is able to recreate the more complex, context-dependent actions that the expert trained it to do. In experiments, when users controlled the robot with this “latent action” algorithm alone, they could pick up an egg, an apple, and a cup of flour and drop them in a bowl (making an “apple pie,” so to speak) faster than an existing approach that required every individual axis of rotation to be moved, one at a time. Once robots get good enough (like flippy), they can preform entire tasks on their own even with environmental interference and without joystick input.

But that’s not all. Today, there’s hundreds of other robotics artificial intelligence systems tackling thousands of other dynamic robotics problems, so be it smart vision for agricultural robots, or human and canine locomotion in robots like ATLAS or SPOT, our robots are getting better at interacting with our world everyday

But… why does this really really matter? For about a century now we have been talking about the robotic takeover of the world, and while we have robots everywhere nowadays, they really remain within the hands of institutions and firms with enough resources to handle them. By making robots smart, we can make these machines be programmable by anyone, we can empower small businesses with the technology previously available only to major corporations, we can make true caregiving robots for the elderly, or autonomous micro-manufacturing robot to make anything from scarves to dog toys right at your desk.

But it comes at a cost, which is an understanding of how our machines work. Take Flippy’s dimensionality reduction. It works incredibly efficiently, and the outcomes are almost always what is desired, but even the research team that developed this method of machine learning could not discern why the robot did what it did. With that, uncovering machine learning bias will become far more difficult, and poorly trained algorithms could have devastating unforeseen consequences.

The future of how much cobots will benefit or harm our lives in the future may be uncertain, but they are coming.