8 SECONDS LATER

The Story

Based on Korea's national Autonomous Driving AI Challenge, four students enter a competition to build the intelligence that lets a car read the road before it happens.

Across twelve episodes, every breakthrough in how a machine perceives, predicts, and plans mirrors a turning point in their own lives.

Then a single failed simulation — a car that drifts toward a person who was never really there — forces them to ask what it truly means to trust a machine with a human life. Winning was never the point. Learning to predict the future was.

Why It Matters

Entertainment that teaches AI.

No lectures. Every episode hides one real concept from modern AI inside a moment of drama — so an audience finishes the series understanding how autonomous intelligence actually works.

Authentic

Real tech, no fiction

Actual competition metrics and baseline models — minADE, ScenarioScore, End-to-End, reinforcement learning — become plot, not props.

Emotional

Failure is the story

A car spins in circles; a leaderboard win collapses overnight. Every technical failure is a human one.

Expansive

From roads to people

Trajectory prediction becomes fall-risk prediction. The same AI that drives a car can help protect a life.

The Season · One Episode, One Idea

Twelve turns of the wheel.

The Starting Line

Perception → Prediction → Planning

Four strangers become a team. Self-driving isn't one AI — it's a pipeline that sees, predicts, and decides.

A World of Data

Public datasets · train / test split

260,000 scenes and no GPU. Great AI begins with great data — and the humility to be dwarfed by it.

To See

3D object detection & tracking

To "see" is to turn pixels into 3D boxes and motion. One mislabeled frame can break everything after it.

One Second Back, Eight Seconds Ahead

Motion prediction · minADE · SMART

From one second of the past, predict eight seconds of the future — the title, and the heart, of the series.

Do Not Collide

Motion planning · ScenarioScore

Safety, progress, comfort — a good decision is never one goal, but the balance of many.

The Reward Trap

Reinforcement learning · reward hacking

The car circles in place. An AI does not do what you say — it does what you reward.

Six Eyes

End-to-End · sensor fusion · VAD

Six cameras, one decision. End-to-End wires the eyes straight to the wheel — and reopens an old wound.

0.028 Seconds

Real-time cut-off · accuracy vs speed

Accurate but slow, or fast but wrong. In the real world, intelligence runs against a clock.

Ghost on the Leaderboard

Generalization · overfitting

Chase the visible score and you lose the real one — how a model performs on a world it has never seen.

Off the Road

Failure scenarios · AI ethics

A collision is not minus points. It is a person. The team nearly breaks apart.

Reproduce

Reproducibility · honest science

A lucky score is not science. What you cannot rebuild, you never really had.

The Ones Who Predict the Future

Technology transfer · healthcare

Beyond the finish line, the same intelligence reaches from the road to the hospital — to protect people.

The Team

Four reasons to predict the future.

LEE SEO-JUN

Team Lead · End-to-End

Lost his brother to a car crash as a child. "If the car had stopped itself." The question that drives him — and the trauma he must face in six camera feeds.

PARK HA-NEUL

Prediction · AI Healthcare

After her grandmother's fall, she became obsessed with knowing the next eight seconds. The heart of the team — determined to move prediction from cars to people.

JEONG WOO-JIN

Planning · Reinforcement Learning

A former game ranker who calls reward design "real-world gaming." The comic spark and the closer — until a badly shaped reward causes a spectacular crash.

KIM MIN-SEO

Perception · Data

The team's eyes, awake through the night over 260,000 labels. "AI is made of data." She catches the auto-label errors that save the team, again and again.

8 SECONDS
LATER.