Diffusion Policy Visuomotor Policy Learning Via Action Diffusion — Paper Explained
The problem statement Diffusion Policy solves is Visuomotor manipulation...
The problem statement Diffusion Policy solves is Visuomotor manipulation, means a imitation learning policy that uses Diffusion Process that performs manipulation tasks given only the camera frames and its current joint state as input, and predicts a set of action chunks as output. Here “an action chunk” means 7 DoF EEF pose ($\text{[x, y, z, roll, pitch, yaw, gripper]}$). The model predicts a set of $n$ (say $n$=16) actions, meaning that if there are 16 future time steps, the end-effector of the manipulator needs to follow a sequence of poses/action chunks to reach the goal position.
[Read More]
