GreedyGeodesicFollower class
Planner that greedily fits actions to follow the geodesic shortest path.
The planner plans on perfect actions (assumes actuation noise is unbiased) and thus
requires the move_forward, turn_left, and turn_right actions to be
present in the agents action space. If you would like to use different actions
(i.e noisy actions), you can override the action key ommited for a given action.
Planner code heavily inspired by https://github.com/s-gupta/map-plan-baseline
Methods
- def find_path(self, goal_pos: numpy.ndarray) -> typing.List[typing.Any]
- Finds the sequence actions that greedily follow the geodesic shortest path from the agent’s current position to get to the goal
- def next_action_along(self, goal_pos: numpy.ndarray) -> typing.Any
- Find the next action to greedily follow the geodesic shortest path from the agent’s current position to get to the goal
- def reset(self) -> None
Special methods
- def __attrs_init__(self, pathfinder: PathFinder, agent: agent.Agent, goal_radius: typing.Optional[float], action_mapping: typing.Dict[GreedyFollowerCodes, typing.Any], impl: GreedyGeodesicFollowerImpl, forward_spec: agent.ActuationSpec, left_spec: agent.ActuationSpec, right_spec: agent.ActuationSpec, last_goal: typing.Optional[numpy.ndarray]) -> None
- Method generated by attrs for class GreedyGeodesicFollower.
- def __init__(self, pathfinder: PathFinder, agent: agent.Agent, goal_radius: typing.Optional[float] = None, *, stop_key: typing.Optional[typing.Any] = None, forward_key: typing.Optional[typing.Any] = None, left_key: typing.Optional[typing.Any] = None, right_key: typing.Optional[typing.Any] = None, fix_thrashing: bool = True, thrashing_threshold: int = 16) -> None
- Constructor
Properties
- pathfinder: PathFinder get set del
- agent: agent.Agent get set del
- goal_radius: typing.Optional[float] get set del
- action_mapping: typing.Dict[GreedyFollowerCodes, typing.Any] get set del
- impl: GreedyGeodesicFollowerImpl get set del
- forward_spec: agent.ActuationSpec get set del
- left_spec: agent.ActuationSpec get set del
- right_spec: agent.ActuationSpec get set del
- last_goal: typing.Optional[numpy.ndarray] get set del
Method documentation
def habitat_sim.
Finds the sequence actions that greedily follow the geodesic shortest path from the agent’s current position to get to the goal
| Parameters | |
|---|---|
| goal_pos | The position of the goal |
| Returns | The list of actions to take. Ends with None. |
This is roughly equivilent to just calling next_action_along() until
it returns None, but is faster.
def habitat_sim.
Find the next action to greedily follow the geodesic shortest path from the agent’s current position to get to the goal
| Parameters | |
|---|---|
| goal_pos | The position of the goal |
| Returns | The action to take |
def habitat_sim.
Constructor
| Parameters | |
|---|---|
| pathfinder | Instance of the pathfinder that has the correct navmesh already loaded |
| agent | Agent to fit actions for. This agent’s current
configuration is used to specify the actions. The fitted actions will
also correspond to keys in the agents action_space. None is used
to signify that the goal location has been reached |
| goal_radius | Specifies how close the agent must get to the goal
in order for it to be considered reached. If None, 0.75
times the agents step size is used. |
| stop_key | The action key to emit when the agent should stop.
Default None |
| forward_key | The action key to emit when the agent should take the move_forward action Default: The key of the action that calls the move_forward actuation spec |
| left_key | The action key to emit when the agent should take the turn_left action Default: The key of the action that calls the turn_left actuation spec |
| right_key | The action key to emit when the agent should take the turn_right action Default: The key of the action that calls the turn_right actuation spec |
| fix_thrashing | Whether or not to attempt to fix thrashing |
| thrashing_threshold | The number of actions in a left -> right -> left -> .. sequence needed to be considered thrashing |