mirror of
https://github.com/Farama-Foundation/Gymnasium.git
synced 2025-08-01 22:11:25 +00:00
e913bc81b8
* feat: add `isort` to `pre-commit` * ci: skip `__init__.py` file for `isort` * ci: make `isort` mandatory in lint pipeline * docs: add a section on Git hooks * ci: check isort diff * fix: isort from master branch * docs: add pre-commit badge * ci: update black + bandit versions * feat: add PR template * refactor: PR template * ci: remove bandit * docs: add Black badge * ci: try to remove all `|| true` statements * ci: remove lint_python job - Remove `lint_python` CI job - Move `pyupgrade` job to `pre-commit` workflow * fix: avoid messing with typing * docs: add a note on running `pre-cpmmit` manually * ci: apply `pre-commit` to the whole codebase
213 lines
11 KiB
Python
213 lines
11 KiB
Python
__credits__ = ["Rushiv Arora"]
|
||
import numpy as np
|
||
|
||
from gym import utils
|
||
from gym.envs.mujoco import mujoco_env
|
||
|
||
DEFAULT_CAMERA_CONFIG = {
|
||
"distance": 4.0,
|
||
}
|
||
|
||
|
||
class HalfCheetahEnv(mujoco_env.MujocoEnv, utils.EzPickle):
|
||
"""
|
||
### Description
|
||
|
||
This environment is based on the work by P. Wawrzy´nski in
|
||
["A Cat-Like Robot Real-Time Learning to Run"](http://staff.elka.pw.edu.pl/~pwawrzyn/pub-s/0812_LSCLRR.pdf).
|
||
The HalfCheetah is a 2-dimensional robot consisting of 9 links and 8
|
||
joints connecting them (including two paws). The goal is to apply a torque
|
||
on the joints to make the cheetah run forward (right) as fast as possible,
|
||
with a positive reward allocated based on the distance moved forward and a
|
||
negative reward allocated for moving backward. The torso and head of the
|
||
cheetah are fixed, and the torque can only be applied on the other 6 joints
|
||
over the front and back thighs (connecting to the torso), shins
|
||
(connecting to the thighs) and feet (connecting to the shins).
|
||
|
||
### Action Space
|
||
The action space is a `Box(-1, 1, (6,), float32)`. An action represents the torques applied between *links*.
|
||
|
||
| Num | Action | Control Min | Control Max | Name (in corresponding XML file) | Joint | Unit |
|
||
|-------|--------------------------------------|---------------|----------------|---------------------------------------|-------|------|
|
||
| 0 | Torque applied on the back thigh rotor | -1 | 1 | bthigh | hinge | torque (N m) |
|
||
| 1 | Torque applied on the back shin rotor | -1 | 1 | bshin | hinge | torque (N m) |
|
||
| 2 | Torque applied on the back foot rotor | -1 | 1 | bfoot | hinge | torque (N m) |
|
||
| 3 | Torque applied on the front thigh rotor| -1 | 1 | fthigh | hinge | torque (N m) |
|
||
| 4 | Torque applied on the front shin rotor | -1 | 1 | fshin | hinge | torque (N m) |
|
||
| 5 | Torque applied on the front foot rotor | -1 | 1 | ffoot | hinge | torque (N m) |
|
||
|
||
### Observation Space
|
||
|
||
Observations consist of positional values of different body parts of the
|
||
cheetah, followed by the velocities of those individual parts (their derivatives) with all the positions ordered before all the velocities.
|
||
|
||
By default, observations do not include the x-coordinate of the cheetah's center of mass. It may
|
||
be included by passing `exclude_current_positions_from_observation=False` during construction.
|
||
In that case, the observation space will have 18 dimensions where the first dimension
|
||
represents the x-coordinate of the cheetah's center of mass.
|
||
Regardless of whether `exclude_current_positions_from_observation` was set to true or false, the x-coordinate
|
||
will be returned in `info` with key `"x_position"`.
|
||
|
||
However, by default, the observation is a `ndarray` with shape `(17,)` where the elements correspond to the following:
|
||
|
||
|
||
| Num | Observation | Min | Max | Name (in corresponding XML file) | Joint| Unit |
|
||
|-----|-----------------------|----------------------|--------------------|----------------------|--------------------|--------------------|
|
||
| 0 | z-coordinate of the front tip | -Inf | Inf | rootz | slide | position (m) |
|
||
| 1 | angle of the front tip | -Inf | Inf | rooty | hinge | angle (rad) |
|
||
| 2 | angle of the second rotor | -Inf | Inf | bthigh | hinge | angle (rad) |
|
||
| 3 | angle of the second rotor | -Inf | Inf | bshin | hinge | angle (rad) |
|
||
| 4 | velocity of the tip along the x-axis | -Inf | Inf | bfoot | hinge | angle (rad) |
|
||
| 5 | velocity of the tip along the y-axis | -Inf | Inf | fthigh | hinge | angle (rad) |
|
||
| 6 | angular velocity of front tip | -Inf | Inf | fshin | hinge | angle (rad) |
|
||
| 7 | angular velocity of second rotor | -Inf | Inf | ffoot | hinge | angle (rad) |
|
||
| 8 | x-coordinate of the front tip | -Inf | Inf | rootx | slide | velocity (m/s) |
|
||
| 9 | y-coordinate of the front tip | -Inf | Inf | rootz | slide | velocity (m/s) |
|
||
| 10 | angle of the front tip | -Inf | Inf | rooty | hinge | angular velocity (rad/s) |
|
||
| 11 | angle of the second rotor | -Inf | Inf | bthigh | hinge | angular velocity (rad/s) |
|
||
| 12 | angle of the second rotor | -Inf | Inf | bshin | hinge | angular velocity (rad/s) |
|
||
| 13 | velocity of the tip along the x-axis | -Inf | Inf | bfoot | hinge | angular velocity (rad/s) |
|
||
| 14 | velocity of the tip along the y-axis | -Inf | Inf | fthigh | hinge |angular velocity (rad/s) |
|
||
| 15 | angular velocity of front tip | -Inf | Inf | fshin | hinge | angular velocity (rad/s) |
|
||
| 16 | angular velocity of second rotor | -Inf | Inf | ffoot | hinge | angular velocity (rad/s) |
|
||
|
||
### Rewards
|
||
The reward consists of two parts:
|
||
- *forward_reward*: A reward of moving forward which is measured
|
||
as *`forward_reward_weight` * (x-coordinate before action - x-coordinate after action)/dt*. *dt* is
|
||
the time between actions and is dependent on the frame_skip parameter
|
||
(fixed to 5), where the frametime is 0.01 - making the
|
||
default *dt = 5 * 0.01 = 0.05*. This reward would be positive if the cheetah
|
||
runs forward (right).
|
||
- *ctrl_cost*: A cost for penalising the cheetah if it takes
|
||
actions that are too large. It is measured as *`ctrl_cost_weight` *
|
||
sum(action<sup>2</sup>)* where *`ctrl_cost_weight`* is a parameter set for the
|
||
control and has a default value of 0.1
|
||
|
||
The total reward returned is ***reward*** *=* *forward_reward - ctrl_cost* and `info` will also contain the individual reward terms
|
||
|
||
### Starting State
|
||
All observations start in state (0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,
|
||
0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0, 0.0,) with a noise added to the
|
||
initial state for stochasticity. As seen before, the first 8 values in the
|
||
state are positional and the last 9 values are velocity. A uniform noise in
|
||
the range of [-`reset_noise_scale`, `reset_noise_scale`] is added to the positional values while a standard
|
||
normal noise with a mean of 0 and standard deviation of `reset_noise_scale` is added to the
|
||
initial velocity values of all zeros.
|
||
|
||
### Episode Termination
|
||
The episode terminates when the episode length is greater than 1000.
|
||
|
||
### Arguments
|
||
|
||
No additional arguments are currently supported in v2 and lower.
|
||
|
||
```
|
||
env = gym.make('HalfCheetah-v2')
|
||
```
|
||
|
||
v3 and beyond take gym.make kwargs such as xml_file, ctrl_cost_weight, reset_noise_scale etc.
|
||
|
||
```
|
||
env = gym.make('HalfCheetah-v3', ctrl_cost_weight=0.1, ....)
|
||
```
|
||
|
||
| Parameter | Type | Default |Description |
|
||
|-------------------------|------------|----------------------|-------------------------------|
|
||
| `xml_file` | **str** | `"half_cheetah.xml"` | Path to a MuJoCo model |
|
||
| `forward_reward_weight` | **float** | `1.0` | Weight for *forward_reward* term (see section on reward) |
|
||
| `ctrl_cost_weight` | **float** | `0.1` | Weight for *ctrl_cost* weight (see section on reward) |
|
||
| `reset_noise_scale` | **float** | `0.1` | Scale of random perturbations of initial position and velocity (see section on Starting State) |
|
||
| `exclude_current_positions_from_observation`| **bool** | `True` | Whether or not to omit the x-coordinate from observations. Excluding the position can serve as an inductive bias to induce position-agnostic behavior in policies |
|
||
|
||
|
||
### Version History
|
||
|
||
* v3: support for gym.make kwargs such as xml_file, ctrl_cost_weight, reset_noise_scale etc. rgb rendering comes from tracking camera (so agent does not run away from screen)
|
||
* v2: All continuous control environments now use mujoco_py >= 1.50
|
||
* v1: max_time_steps raised to 1000 for robot based tasks. Added reward_threshold to environments.
|
||
* v0: Initial versions release (1.0.0)
|
||
"""
|
||
|
||
def __init__(
|
||
self,
|
||
xml_file="half_cheetah.xml",
|
||
forward_reward_weight=1.0,
|
||
ctrl_cost_weight=0.1,
|
||
reset_noise_scale=0.1,
|
||
exclude_current_positions_from_observation=True,
|
||
):
|
||
utils.EzPickle.__init__(**locals())
|
||
|
||
self._forward_reward_weight = forward_reward_weight
|
||
|
||
self._ctrl_cost_weight = ctrl_cost_weight
|
||
|
||
self._reset_noise_scale = reset_noise_scale
|
||
|
||
self._exclude_current_positions_from_observation = (
|
||
exclude_current_positions_from_observation
|
||
)
|
||
|
||
mujoco_env.MujocoEnv.__init__(self, xml_file, 5)
|
||
|
||
def control_cost(self, action):
|
||
control_cost = self._ctrl_cost_weight * np.sum(np.square(action))
|
||
return control_cost
|
||
|
||
def step(self, action):
|
||
x_position_before = self.sim.data.qpos[0]
|
||
self.do_simulation(action, self.frame_skip)
|
||
x_position_after = self.sim.data.qpos[0]
|
||
x_velocity = (x_position_after - x_position_before) / self.dt
|
||
|
||
ctrl_cost = self.control_cost(action)
|
||
|
||
forward_reward = self._forward_reward_weight * x_velocity
|
||
|
||
observation = self._get_obs()
|
||
reward = forward_reward - ctrl_cost
|
||
done = False
|
||
info = {
|
||
"x_position": x_position_after,
|
||
"x_velocity": x_velocity,
|
||
"reward_run": forward_reward,
|
||
"reward_ctrl": -ctrl_cost,
|
||
}
|
||
|
||
return observation, reward, done, info
|
||
|
||
def _get_obs(self):
|
||
position = self.sim.data.qpos.flat.copy()
|
||
velocity = self.sim.data.qvel.flat.copy()
|
||
|
||
if self._exclude_current_positions_from_observation:
|
||
position = position[1:]
|
||
|
||
observation = np.concatenate((position, velocity)).ravel()
|
||
return observation
|
||
|
||
def reset_model(self):
|
||
noise_low = -self._reset_noise_scale
|
||
noise_high = self._reset_noise_scale
|
||
|
||
qpos = self.init_qpos + self.np_random.uniform(
|
||
low=noise_low, high=noise_high, size=self.model.nq
|
||
)
|
||
qvel = (
|
||
self.init_qvel
|
||
+ self._reset_noise_scale * self.np_random.standard_normal(self.model.nv)
|
||
)
|
||
|
||
self.set_state(qpos, qvel)
|
||
|
||
observation = self._get_obs()
|
||
return observation
|
||
|
||
def viewer_setup(self):
|
||
for key, value in DEFAULT_CAMERA_CONFIG.items():
|
||
if isinstance(value, np.ndarray):
|
||
getattr(self.viewer.cam, key)[:] = value
|
||
else:
|
||
setattr(self.viewer.cam, key, value)
|