Ahti Heinla, co-founder of Starship Technologies and CTO
I see robots every day. I see them speeding down the sidewalks at pedestrian speed, stopping to make sure the road crossing is safe. Sometimes I even catch them talking to pedestrians. It’s a look at the fantasies of those with technology, an admirable land of AI. But this is not a hallucination, not a dream, it is a reality that our team of dedicated auditors has built over the last 5 years; we have brought the future to the present.
A couple of years ago, these robots needed a bit of human help and were accompanied on trips, in a format followed by autonomous car manufacturers, who test their cars in public using “safety drivers”.
Starship was the first robotics team to start operating regularly in public spaces about 18 months ago, without the use of a safety guide; we let our robots explore the world on their own. Today, we operate our robot network every day in various cities around the world, bringing people dinner, packages and food.
Shared knowledge is the knowledge gained
It’s exciting to be the first.
When I was a creative engineer on Skype, we were the first to make Voice over IP available in a practical way; now we are working to do the same with robots in public spaces. For four years, our engineering teams have worked behind closed doors in significant progress and amazing experience.
I would like to share with you some details of our technical career. In the coming weeks and months, other members of the Starship engineering team will also share aspects of their journey.
In this course we have dealt with topics such as computer vision, road planning and obstacle detection, which are well-researched in the field of academic robotics. In fact, Starship began as a research project, but soon moved on to a practical delivery operation.
This means that in addition to adjusting the Levenberg-Marquardt algorithm for nonlinear optimization, we had to develop software:
- Automatically calibrate most of our sensors – after all, we don’t want to spend hours calibrating them manually; we have manufactured hundreds of robots and are currently preparing for a larger scale operation.
- Predict how much energy each trip will take from a robot’s battery; so we can orchestrate which robots we will send, depending on their battery status.
- Predict how many minutes it takes a restaurant to prepare food; so the robot will show up on time!
Most autonomous robots in the world today are expensive, built as technology demonstrators or research vehicles, and are not used for commercial operations. A sensor package for a standalone device can cost more than $ 10,000. This just doesn’t work in the shipping space, it’s not a luxury industry where you can charge a premium.
Autonomous driving research vehicles have a computing power of 3 kilowatts in the trunk; It is not practical for a small and safe delivery robot. Therefore, part of our engineering journey has been to design for a lower economic unit. Here are some of the issues we had to consider:
- Advanced image processing on a low-level computing platform.
- Work on software hardware issues.
- Track the frequency of maintenance of robots and why they need it.
- Develop advanced route planning systems to ensure that we are using our robot network effectively.
It has also been a great journey in visual design, with hundreds of sketches, drawings and surveys before the manufacture of our robot’s first plastic body.
When we were still hidden for the first few days, we didn’t want to reveal what our robots looked like. The usual public tests required the creative use of the garbage bag, glued to the robot’s body as a costume!
Building practical robotics is a mix of science, systematic engineering and hackery. This mix of multiple disciplines is a major feature of Starship. Nothing is ever easy in robotics. All knowledge of the situation is probabilistic; all sensors have failure modes and errors, as well as a seemingly easy task stopping the robot at obstacles it can become his small research project.
Starship is a fast-moving startup company and it’s important to turn it into just a big research project. The engineers who are excited about Starship are often not mere scientists, nor mere hackers, nor mere engineers; they have several of these characteristics and can use them as appropriate to the task at hand. We need complicated technical solutions to implement quickly and within the limits of low-cost hardware resources.
Invention and skill are skills that are valued.
It’s been a long time in Starship
Earlier this week, our team will implement a new algorithm to detect point cloud boundaries and test against a whole database of test cases overnight. They will test live on our private test Week.
He will be on the street next Monday as the team will report on our progress at our Monday Engineering Meeting. On most Mondays, some of the engineering teams are making a profit of more than 300% on at least one of the measures achieved in the previous week.
Data as a result of scale and router
Metrics and data have become a big part of Starship engineering.
You see, when we first started we didn’t have data, we still didn’t drive a lot. We changed our robot every day (yes, only that one at the time), took it to the sidewalks and saw how it worked. We now have many, driving autonomously every day, too many for direct observation by engineers.
Thanks to the data, we can now see how our robots work, in the hundreds. We can organize weekly “data immersion” seminars where engineers share findings and watch random submissions to keep in touch with their work.
When we are working to drive our robots faster, we analyze the data in the ‘acceleration events’ table in our Data Warehouse; there are at least 1 billion rows in that table. Other tables include ‘road crossing events’, our maps, all the commands each robot has ever received from our servers, and of course the data received in each delivery they make.
Four years ago, we didn’t have that. When we were just starting out, and we still weren’t doing commercial shipments, I often had to convince people that robotic delivery really works. It was hard for people to believe and he quickly expressed different reasons.
Do skepticism and fear always go with new technologies?
A few years ago, I landed at JFK Airport in New York with a robot in my luggage. Customs, of course, asked, “What is this thing?” I explained that it was a sidewalk delivery robot, and he replied, “Txo, this is New York! They’ll steal it in a minute! ‘
In fact, at the time almost everyone thought they would steal these robots; I’m sure they will (they steal vans to send mail, though rarely). So far our robots have covered more than 200,000 km (130,000 miles) and we have yet to see this problem.
Of course there are security features. The robot has a siren and 10 cameras, is constantly connected to the Internet and knows its exact location with an accuracy of 2 cm (thanks to the aforementioned Levenberg-Marquardt algorithm and 66,000 lines of automatically generated C ++ code, which allows our robots). to use).
People also thought that pedestrians were afraid of robots on the sidewalk or would not accept their presence. Will people call the police? To be honest, we weren’t sure either! However, once we put one of the robots there on the sidewalk, we were in for a big surprise.
Then we were shocked by what happened: people ignored him. The majority of the public didn’t pay any attention to the robots, even those who saw them for the first time, and people certainly weren’t scared. Others would pull out their phone and post on Instagram how they saw the future.
And that’s what we wanted.
We want people to pay as much attention to our robots as they do to dishwashers. This pattern of silently accepting robots as if they were always with us has been repeated in every city in the world in which we operate.
It improves. Once people know that these robots provide a valuable service to the neighborhood, they develop a membership in them. Kids also write thank you letters to robots, we have a ‘thank you wall’ to prove that!
Automating last-mile shipping would never be easy, and we knew it would be a bold project. We also knew that there would always be more than one basic obstacle that needed to be fixed, that there were hundreds of roadblocks! But we have long realized that all these problems can be solved; they just needed ingenuity and perseverance.
Some startups start out running like a sprint, collecting a Minimum Viable Product in 3 months. It’s more of a marathon for Starship – it takes a lot of consistent effort, but the end result brings great benefits to the world.
Delivery is one of the industries in the world that has had a small technological disruption since the last car delivery was approved. The Starship team is working to change that, and with over 20,000 submissions, we’re on our way.
If you’re interested in learning more, check out our second post on the Engineering blog on Neural Networks and how they feed our robots here – https://medium.com/starshiptechnologies/how-neural-networks-power-robots-at-starship-3262cd317ec0