Why Robot Developers Should Embrace the Robotics Edge
September 23, 2019      

For almost a year, robotics developers and companies have used the AWS RoboMaker set of services offered by Amazon as way to develop, simulate and deploy robotics applications through cloud services. Roger Barga, general manager of the service, will be speaking at RoboBusiness 2019 on how companies and developers can take advantage of the service, not just when developing their initial robots, but also when looking for new ways to use their in-production robots.

Barga’s session, titled “The Robotic Edge“, will take place on Wednesday, Oct. 2, in Santa Clara, Calif. Robotics Business Review, which produces RoboBusiness, spoke with Barga ahead of the show about his RoboMaker journey over the past year, surprises from customers, and how companies can create more value for their robots through using cloud services.

DevOps for robotics

Q: What are some of the main themes you plan on sharing with attendees of RoboBusiness?

Barga: The audience will understand robotics and automation, but they’re probably going to be asking the question, ‘What the heck is AWS RoboMaker and how should I think about it?’ They’ll also likely be asking the question of what role does the cloud have to play in robotics, because most software runs on the robot, so ‘Why would I ever want to use the cloud?’ That’s the story arc I’m going to try to take the audience on in my presentation – explain what the service is and how it can help the developer quickly develop and test the robot.

Roger Barga Amazon AWS RoboMaker

Roger Barga, Amazon Web Services

In addition, we have simulation as a service, so you can test your robots through simulation, and then we will help you operate the robot with over-the-air deployment of the application to a robot out in the field. You can monitor robot and even debug your robot once it’s in the field. We’re going to be adding more services to help companies operate their robots once they’re put into production so that people hopefully say, ‘Oh, I understand what DevOps for robotics means.’

Q: What role does the cloud play for robotics development?

Barga: One of the things we’ve learned from being in the market for a year, and also reflecting on other companies that have been doing that, including Amazon Robotics and how we use robots in our fulfillment centers, is that robots are really a key part of a company’s digital transformation. Robots automate tasks that normally people were doing, but because now they have a device out doing the job in collaboration with humans, they can actually get data about how long it takes to do that, or where is the product on the manufacturing floor right now. I can start asking questions when I have that data – like why did it take me twice as long to pick an item off my fulfillment center floor versus this other item?

Companies start getting insights from this data. But in order to prepare for that, they have to be thinking about what data they could get off the robot, and how to instrument it. They have to now start putting on their IT hat on, and often they’ll find that the cloud services we have actually help them with the challenges of capturing data, collecting it and analyzing it. That’s an ‘Aha!’ moment for some people, to realize that the data coming off this robot potentially has value. Once you have more than one or two running around, being able to control their location, divvy up tasks between them, you start to get more value out of your robot.

That’s the second point I want to give to the audience – even if all your code is running on a physical robot at somebody’s house or building simulation as a service, very mature robotics companies use the simulation as part of their CICD [Continuous Integration Continuous Delivery] systems, to run simulation of an app of their robot in production in a simulated environment, to make sure that the robot is not going to regress or encounter failures as the result of a software update. Hundreds of simulations are often run within a few minutes just to check and make sure the application is not regressing.

Simulation surprises

Q: Is the simulation aspect of RoboMaker something that you expected a lot of people to adopt, or has it been a surprise?

Barga: We did not expect that particular use case, and it was an ‘Aha!’ moment for us. Amazon Robotics will physically test our robots – we have a several hundred-square-foot area that our robots would physically go through an exercise before the application was allowed to be updated. But we were surprised to find how many companies were using software simulation and integrating it by hand with their CICD system. So we’ve worked to make that part of our service now.

But again, it should be an ‘Aha!’ moment for a small startup or a group that’s thinking, ‘Why would you ever want to have simulation-as-a-service in the cloud?’ It’s like, ‘Yeah, it makes perfect sense. I’m going to use a hundred machines to run all these simulations in parallel for five minutes. I don’t want to have a hundred machines sitting in the corner of my office building.’

Q: Amazon is coming up on the one-year anniversary of the release of RoboMaker. Have there been some additional lessons or feedback from customers as they’re using it?

Barga: We knew we were going to add a lot of value by allowing over-the-air updates and we knew we’d add a lot of value by streaming data off the robot to a cloud dashboard, or stream any data off the robot, so we invested there. But the conversation has really broadened with new requirements, new ideas, and customers saying to us, ‘Could you build or add this to your service?’

The appetite for having reinforcement learning and simulation integrated into RoboMaker is very high, and we’re actively working on that with customers. We continue to have dialogue with customers who said they want to use these new technologies to change the tasks that the robots do in production.

For example, one of the things we showed at re:MARS was the ability to deploy over the air a new machine learning model to a robot, or we could flash the robot and re-program it from the ground up to do a new task. We were using a drone as an example, but this is another example where customers are thinking about whether to buy the asset, as it can do a number of tasks. They can add new sensors on it and get data to provide new insights into what’s going out in the field.

Adding value for developers

Q: Can you give an example of how companies are adding additional features to an existing production robot?

Barga: One of our customers is Woodside Energy, a mining company in Australia, and they started using robots to move ground materials from Point A to Point B over long distances that don’t need a human driving out there. They can put the equipment on and schedule a delivery to arrive at another point, where someone picks the hardware up. They realized the robot was driving by some interesting spots in the field, namely where they had oil reservoirs that were feeding machines that were moving materials. They thought they could put a camera on the robot and as it was going by the reservoirs, they could take a picture and build a machine learning model to actually infer what percentage of oil was in the reservoir. If it was below 20%, they could contact the cloud and send an alarm to the dashboard to get somebody out there by the end of the day to fill the oil reservoir. It’s being clever and realizing that they had a data feed they could put sensors on, and then stream the data off it to feed machine learning models.

It’s examples like this that I hope to share with the audience. Robots can give you a lot of insight into your business, but you need to think about how to capture that data and act upon it. You need a lot of other services to build an end-to-end application, but you’ll start to see how together robots with cloud services can help build an end-to-end solution for your business.