A Self-Driving Expert Suddenly Has to Solo-Parent

The Scenario

Newcastle upon Tyne. A late March evening.

The day I saw Helen off at the airport, I walked back to our flat holding Lucky's hand. He didn't quite grasp what "Mummy's gone back to China" meant. He just kept asking: "Will Mummy pick me up tomorrow?"

"Mummy won't be back for a while," I said. "From now on, it's just the two of us."

He nodded, seemingly fine with that. Then he asked a question that had absolutely nothing to do with any of it: "Daddy, why is that bus red?"

From that day on, my life looked like this:

Just after 4 a.m., I'm awake. Not insomnia — a gift from the time zone. At this hour it's midday in Beijing, and my colleagues, students, and collaborators back in China are all online. Review a graduate student's draft paper, follow up on a testing collaboration with a government research centre, check a data delivery plan for my startup — in the two or three hours before Lucky stirs, I juggle three professional identities simultaneously. This quiet window has turned out, unexpectedly, to be my most productive time.

At 7 a.m., Lucky wakes up. Then comes the first battle of the day: getting dressed, brushing teeth, eating breakfast. These three tasks together can take anywhere from 45 minutes to an hour and a half, depending on his "cooperation index" that morning.

At 8:30, I walk him to school. The UK Reception class — a bridge between Chinese kindergarten and Year 1 — starts at 8:45. I watch him disappear through the classroom door, then hurry home.

From 9 a.m. to 3 p.m. is my second block of work time. I'm an associate professor at Jilin University, but I also serve as chief safety expert at an automotive tech company and run a startup that has built one of the world's largest aerial naturalistic driving datasets. Three roles, three jobs, one person. My research field is autonomous driving safety — specifically, figuring out how a self-driving car can operate safely across all kinds of complex, real-world conditions.

At 3 p.m. sharp, I'm back at the school gate. From 3 p.m. until Lucky falls asleep around 8:30, those five and a half hours belong entirely to him. Grocery shopping, cooking, the park, building blocks, excavator videos, bedtime stories, and navigating his "I don't want to brush my teeth" and "I don't want to go to sleep."

After 9 p.m., if I still have the energy, I work another hour or two. But because of a corneal transplant, my doctor has told me repeatedly to limit screen time. So more often than not, I work with headphones on — listening to AI-generated audio summaries of papers, dictating articles by voice.

4 a.m., 9 a.m., 9 p.m. — three fragmented windows of work, squeezed into the gaps between parenting. That's my life in Newcastle. A Chinese father, solo-parenting abroad.

A Strange Discovery

My day job is researching autonomous driving safety. Put simply, I'm trying to answer one question: if a car is going to drive itself, how do we make sure it doesn't crash in all the unexpected situations the real world throws at it?

The core method in this field is called "scenario-driven." We don't write one universal set of rules for every road condition. Instead, we break the real world into specific scenarios — merging onto a motorway in a downpour, a pedestrian darting across the road at night, a sharp bend on an icy surface — and for each one we ask: where's the risk? Where's the safety boundary? How should the system respond?

One evening, after Lucky had fallen asleep, I was sitting on the sofa collecting my thoughts. During the day in the lab, I'd been discussing "how an autonomous driving system judges pedestrian intent at a junction." That evening at home, I'd been navigating "how to judge whether to give in or hold firm when Lucky insists on a chocolate cake at the supermarket."

It hit me: these are the same thing.

Parenting and autonomous driving face the same fundamental problem: in an environment you can never fully predict, an "autonomous system" (child / vehicle) needs to make decisions, and you, as the "safety layer" (parent / supervision system), must judge in each specific scenario: should I intervene? How much? Or should I let them handle it?

The autonomous driving industry has a well-known scale — L0 through L5:

• L0: Full human driving. In parenting: you make every decision. What to wear, what to eat, what to play, who to play with.
• L1: Assisted. The car helps keep the lane, but your hands stay on the wheel. In parenting: you hold the bike while they pedal.
• L2: Partial automation. The car can accelerate, brake, and steer, but you must be ready to take over at any moment. In parenting: they do homework; you sit beside them.
• L3: Conditional automation. In specific scenarios, the car drives itself. In parenting: they walk to the corner shop alone to buy a yoghurt.
• L4: High automation. In most scenarios, the car manages on its own. In parenting: they plan their own weekend.
• L5: Full autonomy. In parenting: they're grown up, making all their own decisions.

Every parent makes "scenario judgments" every day: at what level should I be right now?

Lucky wants to cross the road by himself — L0, you hold his hand tight.
Lucky wants to pour his own milk — L2, you watch, ready to catch the wobbling carton.
Lucky gets into an argument with a stranger's child in the park — L3, you observe from a distance, stepping in only if things escalate.

This isn't a metaphor. It's literally what I do every day in two parallel worlds — one on a computer screen in the lab, the other on the streets and in the kitchen of Newcastle.

The Nice Grandpa

It was during these solo-parenting days that I stumbled upon an old "friend" in the Newcastle city library.

If you've spent any time on Chinese social media, you've seen him. The wrinkly-faced old man giving a thumbs-up and saying "Nice!"

But in the UK, I discovered who he really is. He's Michael Rosen — former Children's Laureate, veteran BBC broadcaster, professor of education. He wrote We're Going on a Bear Hunt, a picture book that has accompanied millions of childhoods around the world.

The book in my hands, though, was something he wrote for parents: Good Ideas: How to Be Your Child's (and Your Own) Best Teacher.

I was hooked from the table of contents. Rosen didn't divide education into "maths," "language," or "science." He divided it by living spaces: the kitchen, the bathroom, the sitting room, the loo, the park, the street, travelling... His core argument is disarmingly simple:

Real learning should not be locked inside a classroom. It should happen in every corner of life.

He gave parents four keywords: Investigation (ask the questions you genuinely want answered), Interpretation (don't just memorise answers — form your own understanding), Invention (make things, don't just watch others make them), and Co-operation (learning is not a solo act — it's something we do together).

And he said one thing that changed me: "When your child asks you a question and you don't know the answer, don't make one up. Don't say 'go and do your homework.' Say: I don't know. Let's try to figure it out together."

I later learned that when Rosen wrote this book, his eldest son Eddie had already died — aged 18, of meningitis. In 2020, Rosen himself caught COVID-19, spent 47 days in a coma, and very nearly didn't make it.

A man who has lived through the death of a child and his own brush with death, yet still greets the world with that "Nice!" grin. He didn't write this book to teach parents how to parent. He wrote it to say: every moment you spend with your child is precious. Don't waste it on anxiety. Use it to investigate, interpret, invent, co-operate.

Why 2.0?

Rosen's ideas resonate deeply with me. But his book was written in 2014.

That was a world without ChatGPT, without algorithmic short-video feeds, without smart home devices.

As a father in 2026 — and specifically as someone who researches AI and autonomous driving — I see two gaps in the original:

A technology gap. Rosen taught children to use Google. Now, when Lucky and I are in the loo debating "Why does China have squat toilets but England doesn't?", we can ask AI directly — and AI doesn't just answer, it sparks questions neither of us had thought of. AI isn't the enemy. It can be the third learning partner in the parent-child relationship.

A cultural gap. Rosen's examples are quintessentially British: visiting castles, studying pub signs. What Chinese families know better is making dumplings at Chinese New Year, the courier locker outside the apartment block, five hours on the high-speed train.

But the most important gap is methodological.

Rosen gave us philosophy — investigate, interpret, invent, co-operate. Brilliant. But he didn't give parents a framework for deciding: in this specific scenario, what do I do? Step in or step back? How much?

That framework happens to be the one I use every day at work: scenario-driven.

In autonomous driving development, we decompose the real world into thousands of scenarios, then design, test, and validate for each one.

Parenting can be thought about in exactly the same way:

Don't learn one "correct parenting theory" and apply it to every moment. Instead, in each concrete scenario — the child asks something you can't answer, it's raining and they're bored, they've taken a toy apart — observe, judge, respond.

Rosen gave us the keywords. The scenario-driven approach from autonomous driving gives us a framework. AI gives us a new tool.

Stack these three layers together, and you get Good Ideas 2.0.

How to Use This Book

This is not a book you need to read front to back.

It's a scenario library. Flip to whichever scenario you're living through right now, and see what Rosen says, what I did, and how AI can help.

Each chapter has three parts:

1. Scenario Story: A real experience with Lucky. No right answers — just one father's honest account.
2. L-Rating: What intervention level did I choose in this scenario? Why? In hindsight, was it right?
3. AI Practice: A prompt you can copy-paste straight into an AI chat and use with your child.

This book is also an open-source project. If you have your own scenario to share, you're welcome to contribute. Every family's scenario library is different — your experience might be exactly what another parent needs.

Ready?

Let's begin with the first scenario.

L-Rating

This chapter's scenario: a self-driving expert suddenly has to solo-parent.

My rating: this scenario was a forced upgrade from L1 to full operational coverage.

Back in China, with Helen, both sets of grandparents, and an established routine, we were "dual-driver" or even "multi-driver." My involvement in parenting sat comfortably at L1 to L2 — I participated occasionally but wasn't the primary operator.

After Helen went home, I became the sole driver. No co-pilot. No safety operator. From school pick-up to lights-out, every micro-decision was mine.

This meant I had to expand my "scenario coverage" from roughly 30% to 100% almost overnight. Things I'd never had to manage (what Lucky wears in the morning, what to cook for lunch, what to do when he has a conflict with a classmate) all entered my "operational design domain."

I'll be honest: the first two weeks were chaotic. But it was precisely that chaos that forced me to start applying my most familiar thinking tool — scenario-based reasoning — to the mess of everyday parenting.

Eventually, I made peace with it: this isn't a disaster. It's a window — my window of one-on-one time with Lucky. Once it closes, it's gone.

AI Practice

Introducing AI to your child:

Prompt: "Hello. My child is 5 years old and his name is Lucky. We've just started a 'Good Ideas' project — exploring interesting questions in everyday life. Lucky is absolutely fascinated by excavators right now. Could you explain how an excavator's bucket works in a way a 5-year-old can understand? If possible, use an analogy from something in his daily life."

Expected outcome: AI will explain hydraulic principles in simple language, perhaps comparing it to squeezing a water gun. The point isn't the answer itself — it's the process of asking together. When your child sees you asking AI a question, what they learn isn't hydraulics. They learn that it's okay not to know, and that the thing to do is ask. This is exactly what Rosen calls the power of "I don't know."

A side note: the book you're reading right now — its GitHub repository, bilingual website, the AI prompts in every chapter — was mostly built after Lucky fell asleep, with the help of AI tools. Without AI, given my current rhythm of three jobs plus solo parenting, this book would probably never exist. AI won't parent for you, but it can help you do things in fragmented time that used to require long, uninterrupted stretches.