sydneyli / probablistic-traffic

Probabalistic traffic simulations

CS269i project

Kerem Goksel, Saachi Jain, Sydney Li

Run make and open index.html to play with stuff. Make lets you compile different models together!

• modelHarness.wppl implements functionality that's common to all models like action choosing(act), simulation(simulate), state guessing for players, utility and player switching.
• modelAnalysis.wppl provides analysis methods, all wrapped in an analyzeState(state) method that visualizes various decisions and outcomes for both players when starting from the given state
• commands.wppl is the final file that lets you specify which states you want to analyze
• traffic.wppl is the basic model file. All models need to implement the following:
  • Constants:
    • defaultDepth: int that defines how many steps into the future the agents look when deciding
    • actExecutions: int that determines how many executions the enumerative inference runs when inferring action choice
    • simulateExecutions: int` that determines how many executions the enumerative inference runs when inferring simulation outcomes
  • Methods:
    • pedPrior(state): function that return a distribution over the pedestrian's action given a state
    • carPrior(state): function that returns a distribution over the car's action given a state
    • transition(state, move, player): function that transitions a given state to the new state given the specified player's action
    • crashed(state): returns true if a crash happened in the given state
    • carReward(state): returns the reward the car gets if this is the terminal state
    • pedReward(state): returns the reward the ped gets if this is the terminal state
    • isTerminal(state): returns whether the state is terminal
    • guessCarState(state): returns the pedestrian's guess of the full state (can return the same state if all information is global, otherwise pedestrian has to randomly guess a specific state given its information)
    • guessPedState(state): returns the car's guess of the full state (can return the same state if all information is global, otherwise car has to randomly guess a specific state given its information)