Agents try to find optimal parking spot to minimize cost.
Decide scenario parameters. Given below is an example set of parameters:
street_length = 20
park_probas = {
0: {
'pexist': 0.25,
'poccupied': 0.2,
},
4: {
'pexist': 0.25,
'poccupied': 0.2,
},
1: {
'pexist': 0.4,
'poccupied': 0.6,
},
2: {
'pexist': 0.07,
'poccupied': 0.9,
},
3: {
'pexist': 0.03,
'poccupied': 0.9,
},
}
park_costs = {
0: 7.,
4: 7.,
1: 5.,
2: 3.,
3: 1.,
}
walk_cost = 0.1
drive_cost = 0.1
katoto_cost = 10.0
random_accident_proba = 1e-6
random_accident_cost = 90.0
allow_goback = True
pkatoto_full = 0.4See Parameters section for parameter details.
from streetv2 import StreetParking
env = StreetParking(
street_length=street_length,
park_probas=park_probas,
park_costs=park_costs,
walk_cost=walk_cost,
drive_cost=drive_cost,
katoto_cost=katoto_cost,
random_accident_proba=random_accident_proba,
random_accident_cost=random_accident_cost,
pkatoto_full=pkatoto_full,
allow_goback=allow_goback,
)Street parking has an API nearly identical to OpenAI environments.
env.reset()
state, reward, done, prob = env.step(2)Transition and reward matrices for the environment can be accessed as follows:
>>> env.transition
>>> env.rewards| Parameter | Type | Explanation |
|---|---|---|
street_length |
int |
Length of street in terms of parking spots |
park_probas |
Dict[int: Dict[str: float]] |
Length of street in terms of parking spots |
park_costs |
Dict[int: float] |
Cost of parking for each unique spot |
walk_cost |
float |
Cost of walking per block after parking |
drive_cost |
float |
Cost of driving per block |
katoto_cost |
float |
Cost of parking at the end of the street |
random_accident_proba |
float |
Probability of a costly accident |
random_accident_cost |
float |
Cost of a random accident |
pkatoto_full |
float |
Probability of parking lot being not empty at the end of the street (valid if only allow_goback=True) |
allow_goback |
bool |
Whether to allow going back to beginning of the street if office parking lot is full (use pkatoto_full to control occupancy of office parking lot) |
An example solution with Value Iteration, Policy Iteration and Q-Learning is
provided in example_usage.py.
Example code requires pymdptoolbox for value iteration.