Pusher
State and action spaces
The 23-value observation is the seven arm joint positions, seven arm velocities, fingertip COM (3), object COM (3), and goal COM (3). Object/goal slide-joint coordinates are deliberately omitted. The default action is seven controls in \([-2,2]\).
Dynamics and reward
For fingertip \(p_f\), object \(p_o\), goal \(p_g\), and control \(u\),
\[ r=-w_n\lVert p_o-p_f\rVert_2-w_d\lVert p_g-p_o\rVert_2-w_u\lVert u\rVert_2^2. \]
The three signed components are returned as reward_near, reward_dist, and reward_ctrl. Reset randomizes the object in the configured task region while the goal remains fixed.
Episode end
There is no success termination. The task truncates at max_episode_steps.
Configuration
| Field | Default | Meaning |
|---|---|---|
xml_file, frame_skip | embedded, 5 | Model and action duration |
reward_near_weight | 0.5 | Fingertip-to-object penalty |
reward_dist_weight | 1 | Object-to-goal penalty |
reward_control_weight | 0.1 | Squared-control penalty |
| observation shape/dtype | (23,), f64 | Descriptive metadata |
max_episode_steps | 100 | Horizon |
Custom XML must contain bodies named object, goal, and tips_arm and preserve the assumed seven arm joints.
Reference: Gymnasium Pusher.