Autonomy at Scale: How MPDM is Transforming the Future of Autonomy

Only a couple of decades ago, autonomous driving was a futuristic concept that most people still connected to sci-fi literature. But engineers have been looking to make autonomous vehicles (AVs) a reality for much longer than that and their efforts have brought us to where we are today. Over the last few years, autonomous driving technology has experienced leaps and bounds in its development. AVs have successfully carried passengers across busy metropolitan streets and rural countryside.

But what makes an AV work? The traditional approach to autonomous driving is “early commitment” decision-making and then there is our Multi-Policy Decision Making (MPDM) system. MPDM takes autonomous technology to a whole new level, but to understand why, you first have to understand the early commitment approach.

What is Early Commitment decision-making?

Early commitment technology requires that any and all driving decisions are determined in advance of ever encountering a scenario. In other words, the software is programmed to recognize specific variables and to react in fixed ways. The logic can easily be broken down into if/then sentences. For example, if a red light is sensed, then the vehicle must stop at a predefined point.

That sounds great, but there are two big disadvantages to this approach:

1. The software cannot reliably react to situations that have not been previously programmed

2. It takes a lot of time, effort and money to build and maintain

The challenge with Early Commitment

Let’s say an AV is coming up to an intersection. It’s following one vehicle, there is an oncoming vehicle wanting to turn left in front of the AV and a pedestrian that wants to cross the street. With early commitment, an engineer needs to program responses to a variety of different scenarios that could happen. Does the vehicle in front slow down a little? Does the left-turning one try to speedily cut in front of the AV? Or does the pedestrian judge that they have enough time to run across before the AV arrives at their position?

While you were reading that, you probably thought of several other possible scenarios. Instead of a left turn, the oncoming vehicle could do a u-turn or the front vehicle slams on its brakes, and, at the same time, an unseen biker pulls up alongside the AV. It could very easily be all of that, none of it or something completely different. The possibilities are limitless but the resources are not. And when the system is programmed to react in a limited number of ways, there is greater potential for the decision to be wrong and result in negative consequences.

Yes, a company could throw more engineers at the problem and collect even more data, but fundamentally, early commitment systems do not scale well. That’s why we developed our proprietary MPDM autonomous driving software.

MPDM takes the cake

Where early commitment systems rely on everything being programmed before the tires hit the road, MPDM takes a least commitment approach through live onboard simulations. MPDM will observe its surroundings and all the possible variables and then wait as long as possible before committing to a specific action. We purposefully designed it to make safe decisions under uncertainty, even when it has never encountered a scenario before.

This is the same kind of deliberate reasoning that we as humans rely on when we are behind the wheel. Yes, we are conditioned to react in certain ways to certain stimuli, like early commitment systems do when they see a red light. But when things get uncertain, we become cautious, slow down and then become more confident and assertive as we observe other people and things committing to their own actions. With MPDM, our vehicles are able to replicate that aspect of human awareness that is so valuable when driving.

After MPDM observes the positions and velocities of every other person and object, it generates a set of plausible actions for all of them (including those that may be unlikely to happen). It then selects one action for each object and plays out what would happen, rejecting any action on its own part that would result in a negative outcome. Now, take that deliberate human reasoning and speed up the thought processes. MPDM is able to evaluate thousands of imagined scenarios every second and continues to do so as everything changes around the vehicle.

So, as the world changes, MPDM is able to immediately respond with the best outcome.

Three reasons to integrate our MPDM AVs into your city

1. MPDM is designed to handle any situation in which it can model human behavior

2. MPDM is designed to choose safe actions even when there is uncertainty

3. MPDM scales much better than other systems and requires fewer resources

We truly believe that MPDM is the future of autonomous driving. By integrating our AV fleets into 10 public transit systems across the U.S. and in Japan, we’ve shown that our technology makes autonomous transportation smooth, safe and reliable.

To take a deeper dive into MPDM and its functions, read our most recent whitepaper on the subject and reach out if you have more questions.