[Bug Report] MuJoCo Ant & Humanoid have wrong "x_position" & "y_position" info

[Kallinteris-Andreas]

Describe the bug

For the Humanoid and Ant the info['x_position'] and info['y_position'] are not the same as the observations when exclude_current_positions_from_observation=False is passed. (which is what the documentation suggests)

In general, I see no unit tests for MuJoCo info

Humanoid

The humanoid returns the center of mass instead of qpos[0:2]

Gymnasium/gymnasium/envs/mujoco/humanoid_v4.py

Line 378 in ae5d888

xy_position_after = mass_center(self.model, self.data)

Gymnasium/gymnasium/envs/mujoco/humanoid_v4.py

Lines 397 to 398 in ae5d888

	"x_position": xy_position_after[0],
	"y_position": xy_position_after[1],

Ant

The humanoid returns the get_body_com('torso')[0:2] instead of qpos[0:2]

Gymnasium/gymnasium/envs/mujoco/ant_v4.py

Line 315 in ae5d888

xy_position_after = self.get_body_com("torso")[:2].copy()

Gymnasium/gymnasium/envs/mujoco/ant_v4.py

Lines 333 to 334 in ae5d888

	"x_position": xy_position_after[0],
	"y_position": xy_position_after[1],

Code example

import gymnasium as gym
import pytest


# Note: "HumnanoidStandup-v4" does not have `info`
@pytest.mark.parametrize("env_id", ['Ant-v4', 'Ant-v3', 'HalfCheetah-v4', 'HalfCheetah-v3', 'Hopper-v4', 'Hopper-v3', 'Humanoid-v4', 'Humanoid-v3', 'Swimmer-v4', 'Swimmer-v3', 'Walker2d-v4', 'Walker2d-v3'])
def test_verify_info_x_position(env_id):
    """Asserts that the environment has position[0] == info['x_position']"""
    env = gym.make(env_id, exclude_current_positions_from_observation=False)

    _, _ = env.reset()
    obs, _,  _, _, info = env.step(env.action_space.sample())

    assert(obs[0] == info['x_position'])


# Note: "HumnanoidStandup-v4" does not have `info`
@pytest.mark.parametrize("env_id", ['Ant-v4', 'Ant-v3', 'Humanoid-v4', 'Humanoid-v3', 'Swimmer-v4', 'Swimmer-v3'])
def test_verify_info_y_position(env_id):
    """Asserts that the environment has position[0] == info['y_position']"""
    env = gym.make(env_id, exclude_current_positions_from_observation=False)

    _, _ = env.reset()
    obs, _,  _, _, info = env.step(env.action_space.sample())

    assert(obs[1] == info['y_position'])

System info

$ pip list | grep gym  
gym                           0.26.2
gym-notices                   0.0.8
gymnasium                     0.28.1
gymnasium-notices             0.0.1
gymnasium-robotics            1.2.2

Additional context

['HalfCheetah-v5', 'Hopper-v5', 'HumanoidStandup-v5', 'Swimmer-v5', 'Walker2d-v5'] are not affected

Checklist

• I have checked that there is no similar issue in the repo

[pseudo-rnd-thoughts]

@Kallinteris-Andreas could you make tests and update the code. I think we can make the changes in the v4 environments without a version change due to how infrequently exclude_current_positions_from_observation and info are used

[Kallinteris-Andreas]

I would rather not (by that logic, we should fix v3 too), we should keep v4 as is and fix it in v5.

[Kallinteris-Andreas]

@rodrigodelazcano is there a reason the info was implemented this way, or simply an oversight?

[Kallinteris-Andreas]

The info was giving xpos instead of qpos observations, it will be fixed in v5

if anyone is interested as to why it could have been implemented this way you can read
google-deepmind/mujoco#889