Project · 2025
Rerailing Simulation
Finding the best trajectory to put a rail-road vehicle back on the rails.
- Reinforcement Learning
- A* and RRT trajectory algorithms
- Python
- Pygame
Team
4 people
Duration
2 months
Partner
SNCF
Summary
A dual-mode vehicle drives on both rails and roads, and the whole challenge is getting it onto the rails: finding the optimal trajectory to rerail it in a challenging environment. In collaboration with SNCF, our team of four, all in the AI track, built on two earlier projects: we coupled reinforcement learning with an A* trajectory algorithm and tested it in a Pygame simulation.
My role. I worked on the reinforcement learning algorithm.
Timeline
How it unfolded.
Earlier
First group
Reinforcement learning with a DQN algorithm and a Tkinter simulation. Limits: no modelling of the vehicle, and an alignment that was not perfect.
In pictures
A closer look.
Tools
What I worked with.
Python
The language of the simulation and of the learning agent.
Pygame
To build the simulation, with obstacles and terrain levels.
Reinforcement learning
To learn the rerailing maneuver through trial and error.
A*
The trajectory algorithm coupled with the learning, chosen after comparing it with RRT.
Looking ahead
Why not generate the map randomly at the start of every new simulation?
Next project
Chassis Detection