Environments¶
Mujorax ships with 25 environments from the DeepMind (DM) Control Suite.
This page covers an overview for them listing their respective families and direct links in a table.
Environment families¶
The 25 environments are grouped into 13 families that share physics and bodies. Click a card below for full details on the environments in that family, including the action and observation spaces, rewards, starting state, termination, and a gif preview.
-
Acrobot (2 variants)
Two-link underactuated pendulum.
-
Ball in Cup (1 variant)
Planar ball-and-cup catching task.
-
Cartpole (4 variants)
Cart on a track with a free-rotating pole.
-
Cheetah (1 variant)
Planar bipedal cheetah-style runner.
-
Finger (3 variants)
Two-DoF finger interacting with a free-rotating spinner.
-
Fish (1 variant)
3D free-swimming fish in a water-like medium.
-
Hopper (2 variants)
Planar one-legged hopper.
-
Humanoid (3 variants)
21-DoF humanoid with full upper and lower body actuation.
-
Pendulum (1 variant)
Single-link pendulum on a fixed pivot.
-
Point Mass (1 variant)
Planar point mass actuated independently in x and y.
-
Reacher (2 variants)
Two-link planar arm reaching a randomised target.
-
Swimmer (1 variant)
Six-link planar swimmer in a viscous fluid.
-
Walker (3 variants)
Planar bipedal walker.
All environments¶
| Canonical ID | Family | Description |
|---|---|---|
mjx/acrobot_swingup-v0 |
Acrobot | Two-link underactuated pendulum; dense reward for swinging the tip to target |
mjx/acrobot_swingup_sparse-v0 |
Acrobot | Same as acrobot_swingup with a sparse (binary) reward |
mjx/ball_in_cup-v0 |
Ball in Cup | Planar ball-and-cup catching task; sparse reward when caught |
mjx/cartpole_balance-v0 |
Cartpole | Cart starts near upright; dense reward for keeping the pole upright |
mjx/cartpole_balance_sparse-v0 |
Cartpole | Same as cartpole_balance with a sparse reward |
mjx/cartpole_swingup-v0 |
Cartpole | Cart starts hanging; dense reward for swinging up and balancing |
mjx/cartpole_swingup_sparse-v0 |
Cartpole | Same as cartpole_swingup with a sparse reward |
mjx/cheetah_run-v0 |
Cheetah | Planar bipedal cheetah; dense reward proportional to forward speed |
mjx/finger_spin-v0 |
Finger | Two-DoF finger spinning a free body; dense reward for angular velocity |
mjx/finger_turn_easy-v0 |
Finger | Two-DoF finger rotating a body to a target with large tolerance |
mjx/finger_turn_hard-v0 |
Finger | Same as finger_turn_easy with a tighter tolerance |
mjx/fish_swim-v0 |
Fish | 3D free-swimming fish; dense reward for swimming to a randomised target |
mjx/hopper_hop-v0 |
Hopper | One-legged planar hopper; dense reward for forward speed |
mjx/hopper_stand-v0 |
Hopper | One-legged hopper; dense reward for standing upright |
mjx/humanoid_run-v0 |
Humanoid | 21-DoF humanoid; dense reward for matching a running speed |
mjx/humanoid_stand-v0 |
Humanoid | 21-DoF humanoid; dense reward for standing upright |
mjx/humanoid_walk-v0 |
Humanoid | 21-DoF humanoid; dense reward for matching a walking speed |
mjx/pendulum_swingup-v0 |
Pendulum | Single-link pendulum; dense reward for swinging up and balancing |
mjx/point_mass-v0 |
Point Mass | Planar point mass actuated in 2D; dense reward to a randomised target |
mjx/reacher_easy-v0 |
Reacher | Two-link planar arm reaching a large target |
mjx/reacher_hard-v0 |
Reacher | Same as reacher_easy with a smaller target |
mjx/swimmer_swimmer6-v0 |
Swimmer | Six-link planar swimmer; dense reward for the head reaching a target |
mjx/walker_run-v0 |
Walker | Planar bipedal walker; dense reward for running speed |
mjx/walker_stand-v0 |
Walker | Planar bipedal walker; dense reward for standing upright |
mjx/walker_walk-v0 |
Walker | Planar bipedal walker; dense reward for walking speed |
All physics implementations come from MuJoCo Playground's DM Control Suite []. Mujorax simply adds a thin wrapper that exposes them through Envrax's JaxEnv API; no underlying environment logic is modified from the original.