Point Mass

A simple planar point mass that can be actuated independently in x and y directions.

PointMass

Property	Value
Canonical ID	mjx/point_mass-v0
Action space	Box(-1.0, 1.0, (2,), float32)
Observation space	Box(-inf, inf, (4,), float32)
Episode length	1000
Config	{"ctrl_dt": 0.02, "sim_dt": 0.02, "naconmax": 0, "njmax": 5}

Description

The point mass moves to a randomised target position in the 2D plane. It's the simplest physics in the suite — no joints, no contacts, no orientation — and is the natural choice for sanity-checking algorithms before throwing them at anything more complex.

Rewards

Uses a dense reward that multiplies a "near target" tolerance with a small-control penalty:

near_target   = tolerance(
    distance(point_mass, target),
    bounds=(0, target_size),
    margin=target_size,
)
small_control = (4 + tolerance(action, margin=1, sigmoid="quadratic").mean()) / 5
reward = near_target * small_control

The two terms each capture a separate concern:

• near_target — how close the mass sits to the target, smoothly decaying with distance.
• small_control — a soft penalty on action magnitude. The (4 + ...) / 5 mapping keeps it in [0.8, 1.0] so it lightly modulates the primary near_target term rather than dominating it.

Starting state

obs = [-0.2373 -0.0799  0.      0.    ]

(point-mass position followed by velocity — mass initialised at a randomised position with zero velocity.)

Termination

Episode ends when step >= max_steps (default 1000). No early termination.

Usage

import envrax
env = envrax.make("mjx/point_mass-v0")

Reference

Upstream: mujoco_playground/_src/dm_control_suite/point_mass.py.