wam

此内容尚不支持你的语言。

World-action models: predicted camera views and motion for an instruction.

Layer: learned
Needs: world_model
Requires: –
Conflicts with: –
Never dropped: no
Mixin: skills.wam:WorldModels
Manifest: skills/wam/skill.toml

Functions

`predict_future`

def predict_future(name, instruction, horizon=None)

Ask a world-action model what happens if the robot follows instruction from the current state. Nothing moves. Returns a Prediction with .frames (predicted camera images, RGB arrays, oldest first), .actions (predicted end-effector targets per 25 Hz step: {“left_ee_pose”, “left_ee_joint_state”, “right_ee_pose”, “right_ee_joint_state”}, link6 poses in the env frame, grippers 0 closed .. 1 open) and .seconds. horizon keeps the first horizon action steps and the frames up to that time.

`list_world_models`

def list_world_models()

World-action models available to predict_future.

What the model reads

From the RoboDojo prompt pack (prompts/robodojo/), with every skill and feature on:

  predict_future(name, instruction, horizon=None) -> Prediction
      Asks a world-action model what would happen if the robot followed the
      instruction from now; nothing moves. .frames are predicted camera images
      (show() them to look), .actions the predicted end-effector targets per 25 Hz
      step ({"left_ee_pose", "left_ee_joint_state", "right_ee_pose",
      "right_ee_joint_state"}, link6 poses in the env frame). Takes about a minute;
      use it to compare candidate instructions before committing to one.
  list_world_models() -> [{"name", "description"}]