invertedai.api.initialize(location: str, agent_attributes: Optional[List[AgentAttributes]] = None, states_history: Optional[List[List[AgentState]]] = None, traffic_light_state_history: Optional[List[Dict[int, TrafficLightState]]] = None, get_birdview: bool = False, location_of_interest: Optional[Tuple[float, float]] = None, get_infractions: bool = False, agent_count: Optional[int] = None, random_seed: Optional[int] = None, model_version: Optional[str] = None) InitializeResponse[source]#

Initializes a simulation in a given location, using a combination of pre-defined and sampled agents. Pre-defined agents are those present in states_history, and the remaining ones are sampled conditionally on them. For each agent, pre-defined or sampled, there should be one element in agent_attributes. The pre-defined agents should have fully specified static attributes including agent type. The sampled agents can have their agent attributes be either empty or contains only agent type. The agent_count argument is only for backwards compatibility, and setting it is equivalent to padding the agent_attributes list to that length with default AgentAttributes values. If traffic lights are present in the scene, for best results their state should be specified for the current time, and all historical time steps for which states_history is provided. It is legal to omit the traffic light state specification, but the scene will be initialized as if the traffic lights were disabled. Every simulation needs to start with a call to this function in order to obtain correct recurrent states for drive().

  • location – Location name in IAI format.

  • agent_attributes – Static attributes for all agents. The pre-defined agents should be specified first, followed by the sampled agents.

  • states_history – History of pre-defined agent states - the outer list is over time and the inner over agents, in chronological order, i.e., index 0 is the oldest state and index -1 is the current state. The order of agents should be the same as in agent_attributes. For best results, provide at least 10 historical states for each agent.

  • traffic_light_state_history – History of traffic light states - the list is over time, in chronological order, i.e. the last element is the current state. Not specifying traffic light state is equivalent to disabling traffic lights.

  • location_of_interest – Optional coordinates for spawning agents with the given location as center instead of the default map center

  • get_birdview – If True, a birdview image will be returned representing the current world. Note this will significantly impact on the latency.

  • get_infractions – If True, infraction metrics will be returned for each agent.

  • agent_count – Deprecated. Equivalent to padding the agent_attributes list to this length with default AgentAttributes.

  • random_seed – Controls the stochastic aspects of initialization for reproducibility.

  • model_version – Optionally specify the version of the model. If None is passed which is by default, the best model will be used.

class invertedai.api.InitializeResponse(*, recurrent_states: List[Optional[RecurrentState]], agent_states: List[Optional[AgentState]], agent_attributes: List[Optional[AgentAttributes]], birdview: Optional[Image] = None, infractions: Optional[List[InfractionIndicators]] = None, model_version: str)[source]#

Response returned from an API call to iai.initialize().

agent_attributes: List[Optional[AgentAttributes]]#

Static attributes of all initialized agents.

agent_states: List[Optional[AgentState]]#

Initial states of all initialized agents.

birdview: Optional[Image]#

If get_birdview was set, this contains the resulting image.

infractions: Optional[List[InfractionIndicators]]#

If get_infractions was set, they are returned here.

model_version: str#
recurrent_states: List[Optional[RecurrentState]]#

To pass to at the first time step.