Pusher

This Environment is part of MaMuJoCo environments. Please read that page first for general information. The task is Gymansium’s MuJoCo/Pusher.

Action Space

The shape of the action space depends on the partitioning. The partitioning has the following form:

if partitioning is None:

Instantiate	env = mamujoco_v1.parallel_env("Pusher", None)
Agents	agents= ['agent_0']
Number of Agents	1
Action Spaces	{'agent_0' : Box(-1, 1, (7,), float32)}
Part partition	[(r_shoulder_pan_joint, r_shoulder_lift_joint, r_upper_arm_roll_joint, r_elbow_flex_joint, r_forearm_roll_joint, r_wrist_flex_joint, r_wrist_roll_joint,),]

If partitioning, is None then the environment contains a single agent with the same action space as Gymansium’s MuJoCo/Pusher.

Num	Action	Control Min	Control Max	Name (in corresponding XML file)	Joint	Unit
0	Rotation of the panning the shoulder	-2	2	r_shoulder_pan_joint	hinge	torque (N m)
1	Rotation of the shoulder lifting joint	-2	2	r_shoulder_lift_joint	hinge	torque (N m)
2	Rotation of the shoulder rolling joint	-2	2	r_upper_arm_roll_joint	hinge	torque (N m)
3	Rotation of hinge joint that flexed the elbow	-2	2	r_elbow_flex_joint	hinge	torque (N m)
4	Rotation of hinge that rolls the forearm	-2	2	r_forearm_roll_joint	hinge	torque (N m)
5	Rotation of flexing the wrist	-2	2	r_wrist_flex_joint	hinge	torque (N m)
6	Rotation of rolling the wrist	-2	2	r_wrist_roll_joint	hinge	torque (N m)

if partitioning == “3p”:

Instantiate	env = mamujoco_v1.parallel_env("Pusher", "3p")
Agents	agents= ['agent_0', 'agent_1']
Number of Agents	3
Action Spaces	{'agent_0': Box(-1, 1, (3,), float32), 'agent_1' : Box(-1, 1, (1,), float32), 'agent_2': Box(-1, 1, (1,), float32)}
Part partition	[(r_shoulder_pan_joint, r_shoulder_lift_joint, r_upper_arm_roll_joint,), (r_elbow_flex_joint,), (r_forearm_roll_joint, r_wrist_flex_joint, r_wrist_roll_joint),]

The environment is partitioned in 3 parts, one part corresponding to the shoulder, one part corresponding to the elbow, and one part to the wrist.

Agent 0 action space (Shoulder)

Num	Action	Control Min	Control Max	Name (in corresponding XML file)	Joint	Unit
0	Rotation of the panning the shoulder	-2	2	r_shoulder_pan_joint	hinge	torque (N m)
1	Rotation of the shoulder lifting joint	-2	2	r_shoulder_lift_joint	hinge	torque (N m)
2	Rotation of the shoulder rolling joint	-2	2	r_upper_arm_roll_joint	hinge	torque (N m)

Agent 1 action space (Elbow)

Num	Action	Control Min	Control Max	Name (in corresponding XML file)	Joint	Unit
0	Rotation of hinge joint that flexed the elbow	-2	2	r_elbow_flex_joint	hinge	torque (N m)

Agent 2 action space (Wrist)

Num	Action	Control Min	Control Max	Name (in corresponding XML file)	Joint	Unit
0	Rotation of hinge that rolls the forearm	-2	2	r_forearm_roll_joint	hinge	torque (N m)
1	Rotation of flexing the wrist	-2	2	r_wrist_flex_joint	hinge	torque (N m)
2	Rotation of rolling the wrist	-2	2	r_wrist_roll_joint	hinge	torque (N m)

Observation Space

Observation Categories
Default local_categories	[["qpos", "qvel"], ["qpos"]]
Default global_categories	("qpos", "qvel")
Supported observation categories	"qpos", "qvel"

Besides the local observation of each agent (which depend on their parts of the agent, the observation categories and the observation depth), each agent also observes the position of the pusher’s object and the position of the goal. See more at the Gymnasium’s Pusher.

Rewards

All agents receive the same Gymnasium’s Pusher reward.

Starting state

The starting state of the environment is the same as Gymnasium’s Pusher.

Episode End

All agent terminate and truncate at same time given the same conditions as Gymnasium’s Pusher.

Version History

• v1:

  • Now based on Gymnasium/MuJoCo-v5 instead of Gymnasium/MuJoCo-v4 (https://github.com/Farama-Foundation/Gymnasium/pull/572).

• v0: Initial version release, uses Gymnasium.MuJoCo-v4, first implemented here.