Menu
December 23, 2019 | By

AI IRL Podcast Episode 44: The Intersection of AI & Manufacturing


Subscribe via iTunes, Spotify and more.

Robots have carried out manufacturing tasks for quite sometime — but robots on production lines usually require a high amount of inputs on the front end. Plus, once those inputs are created, the robots are “stuck” — throw off the temperature of the manufacturing floor, the orientation of an object, or any other small detail, and that robot has very little ability to respond.

But now, with advanced robotics, at least one company is trying to create more aware robots that can respond to their environment.

On a recent episode of the AI: IRL podcast apple, special guest Costas Boulis (Chief Scientist at Bright Machines) joined us to explain why AI and robotics are so difficult within manufacturing, and what they’re doing at Bright Machines to reinvent manufacturing with advanced robotics and AI.

Costas is the Chief Scientist at Bright Machines, and was previously an Applied Scientist at Amazon and a Data Scientist for Microsoft, among other advanced science positions. He has a PhD in Electrical Engineering, and received scholar positions at both John Hopkins and Stanford.

‘Robots are currently blind’

“Most of the robots in manfucatring are blind, numb, & dumb. They cannot sense the world, they cannot feel the world, & they cannot really see the world.” — Costas Boulis

Most robots in manufacturing are extremely good at precision — once a human has done all the thinking and created an environment that’s extremely repeatable, they can program a robot to do the same task over and over again.

But here’s the problem: shop floors have their own surprises.

  • Lights change
  • Temperatures change
  • Orientation of an object changes

Throw in any one of these surprises, and there’s a chance that robot needs new programming.

At Bright Machines, Costas recognizes this issue, and his goal is to help robots better understand their environment and the world around them.

In manufacturing, both speed of delivery + reaction to environment matter

In manufacturing, it’s a complex environment — there’s a high value of speed of delivery: “How quickly can we get the assembly line up and running?” But, while delivery speed is important, it’s also not a “set it and forget it” when it comes to the assembly line. The environment will change, and the product requirements could also change. So, it’s a twofold approach in AI and manufacturing:

  1. How do we more quickly get the system up and running so it can produce product?
  2. How are we continuing to define, react, and optimize that system over time with a variety of inputs (visual, sensor, data-based, etc.)?

Most AI allows for no variance — meaning you’re starting from scratch every time

Currently, most AI and robotics systems require a complete reboot when moving to a different assembly line. If one customer is creating a product, work has to be done upfront to understand the exact environment where the robot will be working.

Then, even if another assembly line is making a similar product, if the environment is shifted at all, likely, you are starting from scratch, creating and designing the AI within that robot, regardless of how similar the product is.

Unfortunately, the robot essentially “learned” nothing from one location.

Robots operate in much the same way as someone with blindness:

The way a robot sees in the world right now is very similar to how a person with blindness may navigate the world — they seek out edges or other markers, create an anchor point, and then they move x, y, and z points relative to that anchor.

To navigate again, they seek out another anchor point.

Bright Machines is helping robots learn:

Bright Machines envisions a world in which the robots themselves are able to learn from their actions, then respond to what they’ve learned, sharing information across an ecosystem with other robots.

In manufacturing, robotic precision is incredibly important

To derive the point home further, one has to remember the incredible decision of robots required in manufacturing. In another B2B category, it may be sufficient to adopt an AI program (say a martech or HR technology) that does the job of what you need 80-90% accurately. However, in manufacturing, anything short of 100% accuracy can be a major problem.

As one can imagine, the exact location is hugely important in the manufacturing world when it comes to product lines and how robots respond.

You must understand what factors will impact a robot’s performance

The performance of a robot under varying conditions is obviously important, and it’s not always apparent which conditions (temperature, for instance) may impact a robot’s immediate performance, or its longevity on the shop floor.

So, at Bright Machines, they are constantly trying to input new data into their simulation models to understand what impacts the robot and what does not.

What’s next for AI + robots in manufacturing?

It’s somewhat difficult to see over the horizon, but there is certainly work to be done for AI and robots to continue to automate the manufacturing world. But Costas foresees a world in which robots share information, learn from their environment, and respond accordingly. Checkout the progress Costas and his team are making at Bright Machines, and stay tuned on our podcast!

This discussion with Costas Boulis was taken from our AI: In Real Life podcast. If you want to hear more AI episodes like this one, check us out on Apple Podcasts.


Discuss / Read Comments

Leave a Reply

Leave a Reply

Explore our other AI Insights or Recent Posts.

Subscribe to the Bold360 Blog

Add your email to keep up with AI, chatbots, customer engagement tips and more right to your inbox.