planner.core.action_sampler

Module Contents

class planner.core.action_sampler.ActionSampler(plant: jacta.planner.dynamics.simulator_plant.SimulatorPlant, graph: jacta.planner.core.graph.Graph, params: jacta.planner.core.parameter_container.ParameterContainer)

Parameters:

• plant (jacta.planner.dynamics.simulator_plant.SimulatorPlant) –

• graph (jacta.planner.core.graph.Graph) –

• params (jacta.planner.core.parameter_container.ParameterContainer) –

reset() → None

Return type:

None

initialize(plant: jacta.planner.dynamics.simulator_plant.SimulatorPlant, graph: jacta.planner.core.graph.Graph, params: jacta.planner.core.parameter_container.ParameterContainer) → None

Parameters:

• plant (jacta.planner.dynamics.simulator_plant.SimulatorPlant) –

• graph (jacta.planner.core.graph.Graph) –

• params (jacta.planner.core.parameter_container.ParameterContainer) –

Return type:

None

random_directions(node_ids: torch.IntTensor) → torch.FloatTensor

Generate a random direction.

Parameters:

node_ids (torch.IntTensor) –

Return type:

torch.FloatTensor

proximity_directions(node_ids: torch.IntTensor) → torch.FloatTensor

Generate a direction based on the proximity gradient.

Parameters:

node_ids (torch.IntTensor) –

Return type:

torch.FloatTensor

continuation_directions(node_ids: torch.IntTensor) → torch.FloatTensor

Generate same direction as in-edge action.

Parameters:

node_ids (torch.IntTensor) –

Return type:

torch.FloatTensor

gradient_actions(node_ids: torch.IntTensor) → torch.FloatTensor

Generates actions with dynamics gradients. We formulate a quadratic objective From linearized dynamics, the action minimizing the distance to goal is calculated with optimization.

Parameters:

node_ids (torch.IntTensor) –

Return type:

torch.FloatTensor

directions_actions(node_ids: torch.IntTensor, directions: torch.FloatTensor) → torch.FloatTensor

Calculate a set of actions based on sampled directions and the node’s last action

Parameters:

• node – node we are looking to extend

• directions (torch.FloatTensor) – (k, nq) set of directions are looking to expand the node in

• node_ids (torch.IntTensor) –

Returns:

Set of trajectories expanded in the directions

Return type:

torch.FloatTensor

__call__(node_ids: torch.IntTensor) → Tuple[torch.FloatTensor, int, jacta.planner.core.types.ActionType]

Combines each method of sampling to select an action and potentially project it into the eigenspace

Parameters:

node_ids (torch.IntTensor) –

Return type:

Tuple[torch.FloatTensor, int, jacta.planner.core.types.ActionType]