Intersection¶

An intersection negotiation task with dense traffic.

Warning

It’s quite hard to come up with good decentralized behaviors for other agents to avoid each other. Of course, this could be achieved by sophisticated centralized schedulers, or traffic lights, but to keep things simple a rudimentary collision prediction was added in the behaviour of other vehicles.

This simple system sometime fails which results in collisions, blocking the way for the ego-vehicle. I figured it was fine for my own purpose, since it did not happen too often and it’s reasonable to expect the ego-vehicle to simply wait the end of episode in these situations. But I agree that it is not ideal, and I welcome any contribution on that matter.

Update (v1.12): this is due to an oversight in neighbour vehicle detection, check Neighbour vehicles for explanation. Many thanks to @m-walters for reporting this bug and @Lidang-Jiang for proposing a fix.

Usage¶

env = gym.make("intersection-v0")

Versions¶

ID	Description
`intersection-v0`	Initial version. Same-segment neighbour search only. Preserved for reproducibility.
`intersection-v1`	Continuous-action variant. Same neighbour search as `intersection-v0`.
`intersection-v2`	Connected-lane neighbour search enabled by default. Recommended for new experiments.
`intersection-multi-agent-v0`	Multi-agent initial version. Same-segment neighbour search only.
`intersection-multi-agent-v1`	Multi-agent variant with `MultiAgentWrapper`. Same neighbour search as v0.
`intersection-multi-agent-v2`	Multi-agent variant with connected-lane neighbour search enabled.

See Neighbour vehicles for details.

Default configuration¶

{
    "observation": {
        "type": "Kinematics",
        "vehicles_count": 15,
        "features": ["presence", "x", "y", "vx", "vy", "cos_h", "sin_h"],
        "features_range": {
            "x": [-100, 100],
            "y": [-100, 100],
            "vx": [-20, 20],
            "vy": [-20, 20],
        },
        "absolute": True,
        "flatten": False,
        "observe_intentions": False
    },
    "action": {
        "type": "DiscreteMetaAction",
        "longitudinal": False,
        "lateral": True
    },
    "duration": 13,  # [s]
    "destination": "o1",
    "initial_vehicle_count": 10,
    "spawn_probability": 0.6,
    "screen_width": 600,
    "screen_height": 600,
    "centering_position": [0.5, 0.6],
    "scaling": 5.5 * 1.3,
    "collision_reward": IntersectionEnv.COLLISION_REWARD,
    "normalize_reward": False
}

More specifically, it is defined in:

classmethod IntersectionEnv.default_config() → dict[source]¶

Default environment configuration.

Can be overloaded in environment implementations, or by calling configure(). :return: a configuration dict

API¶

class highway_env.envs.intersection_env.IntersectionEnv(config: dict = None, render_mode: str | None = None)[source]¶

classmethod default_config() → dict[source]¶

Default environment configuration.

Can be overloaded in environment implementations, or by calling configure(). :return: a configuration dict

step(action: int) → tuple[ndarray, float, bool, bool, dict][source]¶

Perform an action and step the environment dynamics.

The action is executed by the ego-vehicle, and all other vehicles on the road performs their default behaviour for several simulation timesteps until the next decision making step.

Parameters:: action – the action performed by the ego-vehicle
Returns:: a tuple (observation, reward, terminated, truncated, info)