Reinforcement Learning

Pilot · RL

Pedestrian RL on Shared Streets

Formulating a pedestrian on a sidewalk-less local street (“이면도로”) as an RL agent, testing whether a safety–efficiency reward trade-off produces emergent walking behavior.

A tabular Q-learning agent reaches the goal while keeping distance from the shared-lane vehicle.

With a low safety weight the agent walks straight through the vehicle lane, minimizing travel time.

Raising the safety weight makes a personal-space radius (~2.24) emerge, pushing the path to the road edge.

Transit Dispatch Simulation

A reinforcement-learning agent dispatching vehicles across stations to reduce passenger waiting time.

A reward-shaped gridworld used to train and debug the control policy.