visualizers.viser_app.controllers.sampling_base#

Module Contents#

visualizers.viser_app.controllers.sampling_base.MAX_NUM_TRACES = 5#
class visualizers.viser_app.controllers.sampling_base.SamplingBaseConfig#

Bases: jacta.visualizers.viser_app.controllers.controller.ControllerConfig

Base controller config with spline parameters. .. py:attribute:: horizon

type:

float

value:

1.0

num_nodes: int = 3#
num_rollouts: int = 32#
spline_order: Literal[zero, slinear, cubic] = 'slinear'#
control_freq: float = 20.0#
use_noise_ramp: bool = False#
class visualizers.viser_app.controllers.sampling_base.SamplingBase(task: jacta.visualizers.viser_app.tasks.task.Task, config: SamplingBaseConfig, reward_config: jacta.visualizers.viser_app.tasks.task.TaskConfig)#

Bases: jacta.visualizers.viser_app.controllers.controller.Controller

Base class for all sampling controller implementations. .. py:property:: num_timesteps

type:

int

Helper function to recalculate the number of timesteps for simulation

Parameters:

  • task (jacta.visualizers.viser_app.tasks.task.Task) –

  • config (SamplingBaseConfig) –

  • reward_config (jacta.visualizers.viser_app.tasks.task.TaskConfig) –

property rollout_times: numpy.ndarray#

Helper function to calculate the rollout times based on the horizon length

Return type:

numpy.ndarray

property spline_timesteps: numpy.ndarray#

Helper function to create new timesteps for spline queries.

Return type:

numpy.ndarray

property spline: scipy.interpolate.interp1d#

Spline defining the current control signal to be applied.

Return type:

scipy.interpolate.interp1d

property controls: numpy.ndarray#

Contains the control signals applied in the current rollout.

Return type:

numpy.ndarray

resize_data() → None#

Resize states, sensors, and models to (config.num_rollouts, num_timesteps, …).

Return type:

None

abstract update_action(curr_state: numpy.ndarray, curr_time: float, additional_info: dict[str, Any]) → None#

Abstract method for updating controller actions from current state/time.

Parameters:

  • curr_state (numpy.ndarray) –

  • curr_time (float) –

  • additional_info (dict[str, Any]) –

Return type:

None

abstract action(time: float) → numpy.ndarray#

Abstract method for querying current action from controller.

Parameters:

time (float) –

Return type:

numpy.ndarray

update_spline(times: numpy.ndarray, controls: numpy.ndarray) → None#

Update the spline with new timesteps / controls.

Parameters:

  • times (numpy.ndarray) –

  • controls (numpy.ndarray) –

Return type:

None

set_default_controls() → None#

Set default value for the Controller.controls. if there is no default value set to zero.

Return type:

None

reset() → None#

Reset the controls, candidate controls and the spline to their default values.

Return type:

None

update_traces() → None#

Update traces by extracting data from sensors readings.

We need to have num_spline_points - 1 line segments. Sensors will initially be of shape (num_rollout x num_timesteps * num_physics_substeps x nsensordata) and needs to end up being in shape (num_elite * num_trace_sensors * size of a single rollout x 2 (first and last point of spline) x 3 (3d pos))

Return type:

None

visualizers.viser_app.controllers.sampling_base.make_spline(times: numpy.ndarray, controls: numpy.ndarray, spline_order: str) → scipy.interpolate.interp1d#

Helper function for creating spline objects.

Parameters:

  • times (numpy.ndarray) – array of times for knot points, shape (T,).

  • controls (numpy.ndarray) – (possibly batched) array of controls to interpolate, shape (…, T, m).

  • spline_order (str) – Order to use for interpolation. Same as parameter for scipy.interpolate.interp1d.

  • extrapolate – Whether to allow extrapolation queries. Default true (for re-initialization).

Return type:

scipy.interpolate.interp1d