Is Perceptive Automata a prediction company?
People call Perceptive Automata a prediction company. I don’t like it! Why shouldn’t I like it, isn’t that what we’re doing? It depends how you define prediction. If you asked somebody off the street to define “prediction,” they would probably say something like “describing what will happen in the future”. Not a bad definition! But for autonomous vehicles — and other robots — you need to be more precise. For roboticists, “prediction” means something very specific.
To understand how roboticists think about prediction, you need to understand how they think about navigating the world. For a robot, the first step is to get from point A to point B. To do that, you have to avoid obstacles. When you’re learning to make robots navigate the world, you start with a landscape that has some fixed obstacles. The question you’re trying to answer is what move the robot should make at each step so that it gets to its destination without bumping into anything. The next big step (often years into a course of study in robotics) is to do the same thing, but in a landscape where some of the objects move. This is a much harder problem. In order to solve it, you need prediction. What does prediction mean in that context? It means knowing where the objects will be at a particular point in the future. So if your robot can move one inch per step, and there’s a rolling ball a foot away from the robot that’s moving, where will ball be in 12 robot steps, and how do you avoid it? Then you scale up this world even more. You add more moving objects, and have them move in more complicated ways.
Eventually, you reach the point where your robot is an autonomous car, and the other objects are pedestrians, cyclists, other motorists. In this context, prediction — as you have defined it — is really hard! People don’t move in straight lines, and sometimes they don’t move at all when you think they might. It’s so hard, and people are so unpredictable using traditional methods, that nobody has yet solved this problem well enough to deploy an autonomous vehicle.
These are exactly the kinds of problems that Perceptive Automata’s State of Mind AI is designed to solve. So why don’t I like it when people call us a prediction company? Because the question we start with is not “where will this person be in the future”. It’s “what’s in this person’s head that’s relevant to the task of driving?” The reason we start with that question is that it’s a lot more like the questions that humans are implicitly answering when we drive. When you see a pedestrian, you don’t calculate their trajectory. You look at them. Do they look like they’re trying to cross in front of you or are they waiting for a bus? By asking the question this way — in a human-like way — we’re able to solve the problem of how you interact with humans. This is a problem that roboticists haven’t been able to solve.
When we bring our “State of Mind AI” to our customers to integrate, many times we integrate with their existing prediction system. We help them solve the prediction problem as they’ve defined it. But we do that by asking the question a different way. We have a different mindset, and we use different tools. So when people ask if we’re a human behavior prediction company, I will often say no: we’re a human understanding company.
