We get a lot of questions about how our robots know how and who to follow, so we sat down with Tyson, PFF’s Smart Behaviors Team Lead, and asked him all about the science that goes into our robots. The Smart Behaviors team is responsible for conducting human behavior studies that help inform our robots how to move politely and intuitively in their surroundings. You may have heard us talk about how our robots operate with “pedestrian etiquette” before, or in other words, move in a human-like manner that is dynamic and seamless.

Tyson and the Smart Behaviors team are the brains behind those human-like movements and through their studies, they are able to give information to our engineers who then translate it into a way that gita robots can understand. Here’s what Tyson had to say:

At a high level, what is the process of coming up with a smart behaviors study?

“There are lots of sources that help us determine what the next study should be, and we keep a running list of user experiences that we intend to look into. We generally get input from users, testers, and sales teams on what they would like to see our products do. We also get a lot of ideas from just having conversations with people in our communities. People are really excited when a new product is introduced, and they love sharing their ideas on how they could use the products and what they would do to make it better. With all of these ideas, we prioritize them based on what will be most valuable to the people who use our products every day.”

How long do studies take from start to finish?

“Studies are a heavy lift - they take between 4 and 8 months to go from start to end. We’re looking into totally new territory when it comes to human behaviors, etiquette, and interpersonal relationships, so we have to collect and analyze all of our own data. Nothing comes from outside sources.”

What do you look for when conducting these studies?

“We’re mostly looking for any subtle similarities between all of our participants’ behaviors. From the outside, the studies look fairly simple because in our lab we’re mostly interested in innate behaviors - things we humans do that we don’t even think about when we are doing them. Luckily, our system detects even the smallest nuances in behavior. Shift your weight or slow down just a little bit and we’ll pick it up. If our participants are from diverse backgrounds but they still give the same unconscious signals to the people around them, then we know we’re onto something that will be helpful to users across the board.”

What are the main points of consideration when coming up with a smart behaviors study?

“Bias is always at the top of our minds. We tend to get our initial ideas from others’ input, but when we’re looking at human behaviors, we try to separate ourselves from the limitations of our products and ask, “What are people actually doing out there in the world?” The behaviors we are interested in happen in your subconscious, and that’s a particularly hard thing to study because it can’t be forced. So when we start a study we’re always thinking, “How do we get the most natural, unbiased data we can for this particular behavior?” Better data makes for better robot behaviors.”

How do you deliver results to the team so that they can translate it into software for gita?

“Our main method of delivering results is through a Design Specification: a 20-or-so page document that describes in detail all of the situations in which this robot behavior might arise and what it will look like when it happens. The specification is supplemented with all kinds of cool content. We make detailed drawings and animations that show the ideal robot behaviors in various scenarios. We also use gaming engines to simulate what the behavior will look and feel like in real time. You can toss on a VR headset and walk around with a digital version of our products and get a sense of what it’s like before it gets uploaded onto the robots themselves.”

What is your favorite part about being on the Smart Behaviors team?

“I’m a nerd, and I love facing unprecedented challenges. Our team is small, but we have a super diverse range of skills. We’re a biomechanist, a statistician, an architect, and an anthropologist. When we tap into our expertise to discover something totally new about human behavior, that’s when I get really excited. Then I get to figure out how we can improve the process, quality, and accuracy of our data and our delivery. This bouncing back and forth between discovery and evolution is incredibly gratifying.”