Building robots for unpredictable, infrastructure-free environments

robots for unpredictable, infrastructure-free environments">

Burros can carry, tow, scout, patrol, mow, push, pull, or propel a variety of attachments as a platform for manipulation. | Credit: Burro AI

In 2018, Burro did a demo. The robot worked. We were excited. We thought we understood the problem. We did not understand the problem.

What we understood was how to make a robot perform in conditions we controlled, for an audience prepared to see it succeed, over a time horizon short enough that the long tail of real-world failures hadn’t had time to appear. That is what a demo is. It is a proof of concept for a best-case scenario.

It is not a proof of concept for Tuesday morning in November when it’s raining, and the lighting is flat, and a worker approaches from an unexpected angle, and the robot is operating in a country it has never been to before.

The gap between those two things is where most robotics companies fail. Not because their technology is bad, but because they optimized for the wrong thing for too long. They kept the demo alive while the real-world deployment problem went unsolved.

Burro made a different choice, though not because we were smarter. We made it because we had no alternative. The environments we were working in, outdoor agricultural settings with no fixed infrastructure, no controlled lighting, and no GPS reliability under canopy.

The agricultural workforce was not going to modify its behavior to accommodate a machine. All of these did not permit the kind of controlled-conditions optimization that indoor robotics can sustain for years before hitting the real world. We had to confront the real-world problem immediately, which meant we had to start learning from it immediately.

What we learned first was about tolerance. People who depend on a robot for their livelihood have zero tolerance for unreliability.

When a customer first adopts an autonomous system, they think of it as an interesting new tool. Within weeks, if the system is delivering value, their mental model shifts entirely. They are now depending on it. They have organized their workflow around it. They have told their team to plan around it.

When it fails, they are not mildly disappointed. They are angry in the way you are angry when critical infrastructure fails, because that is what it has become. This shift from novelty to dependency happens faster than most companies expect, and the reliability bar it sets is higher than any lab environment will prepare you for.

What we learned second was about environmental variability. Nothing outdoors is static. The same row looks different at dawn, midday, and dusk. It looks different in summer and winter, in rain and sun, in dust and mud. Temperature ranges from below freezing to 120 degrees Fahrenheit.