Mobility Matters Daily #409 - Automation and Estimation

Good day my good friend.

To start with, thank you so much for your patience over the last week. Finally, everything seems to have cleared, with the help of several walks over the last few days. This time of year is definitely one of the best times to enjoy two feet as a mode of transport. I mean, just look at it.

This was rather nice to walk down

If you have any suggestions for interesting news items or bits of research to include in this newsletter, you can email me.

James

Thank you Landor. Your sponsorship is helping to feed families in need!

Are autonomous vehicles going nowhere?

This article on Bloomberg caused a bit of a stir last week. In short, after having a lot of money thrown at them and no company yet turning a profit, apparently we are nowhere near autonomous driving yet. Needless to say that this has resulted in a lot of self-congratulation on behalf of many transport professionals. Most of whom miss the wider context spectacularly.

Shooting down predictions on driverless tech is as bad as being a dart-throwing chimpanzee - easy to do, and wrong more often than people think. Yes, we are still some way off commercially deployable solutions outside of a few use cases in controlled environments. Autonomous driving is likely to need as near a general purpose artificial intelligence as it is to get without actually being it. And we are WAY off achieving that. But in this game, the vehicle and driving is not the prize: autonomy is. And as more AV companies deploy their autonomous software elsewhere to make money, that in itself will have a profound impact. Just don’t expect Johnny Cab soon.

Yes, I did watch Total Recall last week (Source: Futurelapse)

Estimating is hard, particularly for pollution

Speaking of predictions, it turns out estimating is hard. When you account for uncertainties, in-built assumptions, and biases (no matter how well the data is collected), its a wonder we make predictions at all. We certainly remember the ones that people got right. Anyway. Agent-based modelling is now allowing us to simulate vehicle emissions at the individual level. And when this is scaled up, it turns out we may be underestimating the emissions of vehicles. By a lot.

In a post-Diesel Gate world, underestimating the emissions from vehicles is generally a bad thing. And there is a lot to commend this study for. But there is a word of caution, before stating that an article is good because you agree with the conclusion. Look at the method. Agent-based models are difficult to reproduce, and errors in model calculations can multiply in agent-based simulation. Thankfully, the researchers have published their method openly, and so hopefully someone can take their work and reproduce it.

Thank you Markides Associates. Your sponsorship is helping families in need!

Random things

These links are meant to make you think about the things that affect our world in transport, and not just think about transport itself. I hope that you enjoy them.

• Why does time go forwards, not backwards? (BBC)

• Meet Kastle Systems, the Covid-Era Kings of Back-to-Work Data (Bloomberg)

• Team of flying robots builds structures using 3D printing (Physics World)

• Ukraine’s Starlink problems show the dangers of digital dependency (Financial Times)

• Digital License Plates (Schneier on Security)

Something interesting

The Pacer train is somewhat of a legend here in the UK. And in this Rail Natter with Gareth Dennis, Dr Kevin Tennent says that they were rubbish and did nothing to save threatened lines. You be the judge.

If you do nothing else today, then do this

I will be taking part in one of Glenn Lyons’s Fireside Chats in a couple of weeks. Talking about the role of serious games in policy making, and how we should use them more. Rebecca Posner, Maha Attia, Charles Musselwhite, and Clare Sheffield will be there too. You should sign up.