/controller-synthesis

Primary LanguagePythonMIT LicenseMIT

controller-synthesis

Ideas

  • Constraint as classifier - I

  • Metric(decided by us) for reward shaping based on deviation from constraint to train RL Agent - I

  • Later stage

    • Build an automaton for constraints and use this for deriving metrics - II

Models

  • Model 1

    Cases:

    1. Train is in radius of station
    2. Train is at station
    3. Train is inbetween stations

    Train

    • Case 2 per time delta produce actions, num people in station, train capacity(intrinsic) - Reward function

    • Case 1 speed, and information from station (num people in station) - Reward function

    • Case 3 speed, distance from the upcoming station

    • Action space -> speed, Extrinsic varibles - information from station(num_people, num_trains arriving to the station)

    • Observation space -> num people in the train, train capacity, information of the station it is going towards

    • Transition Function -> path that should be followed by the trains + change in speed

    • Base variables: num_trains, num_stations, sampler function, distances and paths

    • Output: Speed

To-Do

Station class

1. Keep track of num_people and update num_people entered a particular train
2. Keep track of num_trains coming towards the station at the moment
3. Keep track of distances of each train from num_trains

  • Modelling (By Sunday Evening)

    • Train Identifier
    • Station Identifier (Track + people)
    • Action (time)
    • Transition function (fixed)
    • Reward function (Just occupancy at the start and shape it on the fly)
    • gamma factor (1 for now)

  • Metric

    • distance metric (for now, and check rest, what works)

  • RL model (By Saturday evening)

    • RLlib basic framework

  • Constraints (By Saturday evening)

    • Representation
    • Metrics