Let `H` be the length of the trajectory (meaning the number of actions the agent takes). :property actions: Tensor of shape [H, action_d

:propery overweight_freq: How much more frequently is this instruction sampled during training?

Gets the distance between the EE and the closest target object.

Gets the distance between the EE and the closest target object.

Uses inverse kinematics (requires pybullet) to apply end-effector position control for the articulated_agent's arm.

Always outputs the llama tokenization.

For operating over single token inputs.

Reward function for picking an object up.

Always outputs the T5 tokenization.