How AI, ML and Motion Planning Testing Are Advancing New Mobility

February 10, 2022

At the iMerit ML Data Ops summit, Professor John Dolan, Principal Systems scientist from CMU, and Professor Jinhua Zhao, the Urban Mobility lead at MIT, discuss challenges, issues, and glitches on artificial intelligence (AI), and machine learning (ML) and breakthroughs in autonomous vehicles (AVs) with Chris Barker, the CEO of CBC Transportation Consulting.

The discussion focused on how mobility innovation has the potential to dramatically change the way we connect and commute, and how recent developments in machine learning have the potential to enhance safety while altering how we live and move. 

iMerit ML Data Ops summit, Professor John Dolan and Professor Jinhua Zhao

But what factors stand in the way of advancing this mobility?

Assessing the Readiness of Urban Environments and Presence of AVs

While it is common to witness an increase in the number of AVs being launched in cities, the issue remains whether we are prepared for their implementation. The three critical challenges facing us are as follows:

  1. How to make the machine work?
  2. How does the machine fit within the mobility system?
  3. How does the AV affect the city being liveable?

A city prides itself on encouraging social interaction. Hence, the first breakthrough would be to solve the challenge on how an AV must interact with the other components of the system.

The second challenge is really from the mobility system point of view:

  • An autonomous vehicle cannot work alone.
  • It must be a component of a multi-modal system, particularly in the bigger cities, where there is public transit, railways, buses, bicycles, pedestrians.
  • And how would an AV interact with any one of these while being a component to the broader system?

The third challenge is from a policy point of view. The government typically worries about safety, carbon emissions, congestion issues, liveability issues, and equitability issues etc.

Assessing the Readiness of Urban Environments and Presence of AV

Other Mobility Challenges in the AV Ecosystem

The first challenge within the AV ecosystem is how to decarbonize the entire mobility system and the second one is the mobility equity discussion.

Decarbonizing the Entire Mobility System – This is the most significant issue. Mobility is the biggest CO2 emitter in the United States, and electrification innovation is by far the simplest way to decarbonize.

The Mobility Equity Discussion The urban mobility system as of today, is failing in terms of mobility equity. A large chunk of the population for whatever may be the cause, be it disability, lack of income, poor health, or any other reason, do not enjoy enough mobility and accessibility.

The Primary Goal of any AV/ EV Technology

The primary goal of this technology must be to fulfill the goal of access for all, which is the underlying social agenda. However, this would initially be difficult. Like any new technology when it happens, it doesn’t care about equity. Instead, it identifies the most profitable market to survive. And when successful, it penetrates the whole society.

Safety Level Norms

It is difficult to try and get a precise idea of what one means by safety, and then compare that with the public perception of the safety level in the United States.

It is evident that in many of the developing countries, the mobility safety levels are rather poor, but they have a healthier view about the safety of the AV, because people tend to view safety in relative terms, rather than absolute terms. 

If this is the status quo then one has little choice but to embrace it. But, if you are in an environment where the safety situation already has an extremely high standard, then it becomes particularly challenging. For example, the US airline industry has a particularly incredible safety record. 

Now, does the public expect the AV industry to impose the same level of safety as in the US airline industry? The reality is that in the U.S. there are one hundred people killed per day. Hence the social norm balancing point would be to balance between zero tolerance and one hundred, to something which is acceptable for the AV industry.

Safety Level Norms

Biggest Safety Hurdles That Are to be Overcome

The biggest safety hurdle would be to provide a safety level that is comparable, or lower, than the current one with human drivers – it should be less than the current 37,000 deaths per year from accidents in the US, and a million worldwide. 

People expect automation to have a better safety level, and they would want to see guarantees for the same. To have a good safety level, we need to incorporate both AI and ML to successfully operate AVs.

While we are still far from offering absolute AV safety, deep learning has transformed perception in behavior and motion planning, making it an essential component of safe autonomous vehicle driving. 

The AVs with AI capabilities will cover every possible instance, as deep learning has expanded in its scope due to human perception and indecisiveness. In other words, deep learning has widened the decision-making envelope of AI and ML.

The Ideal Notion of ML / AI Decision Making Process System

Researchers and scientists tried to initially introduce a notion of interpretability, and what it represents. They then tried to improvise on this interpretability, so that a user or a system trusts its decision making process.

The first intellection in this case is a generic interpretability discussion on all ML/ AI systems. The second one is robustness. 

Both are quite relevant. However, there’s a lot of new research trying to improve the robustness of the AV, chiefly from a safety point of view. It is based on the robustness of the algorithm.

The Ideal Notion of ML AI Decision Making Process System

Personality of an AV

While deep learning has enhanced the decision making process system, the notion of giving a personality to an AV, is a breakthrough in the AV social intelligence arena, particularly from a safety context. The concept, which was developed by Daniella Roos, CCO at MIT, makes it easy to understand the aggressiveness of the coming vehicle. 

To empower an AV to have this kind of social intelligence would be the foundation of the next generation of AVs.

Final Takeaway

The rapid adoption of AVs has significant implications for the whole mobility value chain. It demonstrates how AI and ML can be used to transform autonomous mobility. 

MIT’s mobility systems are the systems of tomorrow. Companies, cities, and communities can partner with CMU, MIT, and other Universities around the country to improve overall urban mobility.