⚡AI SkyRun Hackathon: Teach an AI to Find a Path It Has Never Seen

🧡 What is this about?

Imagine an agent that knows nothing about the world around it. No map. No hints. No coordinates of the obstacles.

There is only one rule: crash and it hurts, arrive and you get a reward.

And after a few minutes of training, that agent finds a route through a space it has never seen — on its own. Nobody guided it there. It simply tried, failed and remembered the pain.

That is reinforcement learning — the method used to train game bots, robots and control systems. And it is exactly what the hackathon participants will launch with their own hands.

🧡 What will participants have to do?

• Write the learning engine — the thing that actually makes the agent intelligent.
• Decide what the agent gets punished for and what it gets praised for. It sounds simple, but this is where most solutions break: word the reward carelessly and the agent may learn to, say, “stand still safely”.
• Take it from chaotic attempts to a confident route.

🧡 What will you see with your own eyes?

• How the agent’s route changes after every collision, and how its trajectory gradually straightens out.
• How the training results of different teams compare — and why the winner is not the team that tuned its solution to one situation, but the team whose agent genuinely learned.

The finale: the best policy comes alive in a simulator — no longer a diagram, but a full three-dimensional world.

🧡 Who will find this interesting?

Final-year school students who have heard the words “artificial intelligence” and finally want to see what is inside — not on slides, but in code they ran themselves.

The Faculty of Physics and Mathematics is a place where artificial intelligence is studied not as a set of ready-made libraries, but as mathematics: dynamic programming, optimisation, control theory. The AI SkyRun Hackathon is an honest cross-section of what real work looks like here.