Files
Gymnasium/gymnasium/envs/mujoco/walker2d_v5.py
2023-11-05 13:28:17 +00:00

359 lines
20 KiB
Python

__credits__ = ["Kallinteris-Andreas"]
from typing import Dict, Tuple
import numpy as np
from gymnasium import utils
from gymnasium.envs.mujoco import MujocoEnv
from gymnasium.spaces import Box
DEFAULT_CAMERA_CONFIG = {
"trackbodyid": 2,
"distance": 4.0,
"lookat": np.array((0.0, 0.0, 1.15)),
"elevation": -20.0,
}
class Walker2dEnv(MujocoEnv, utils.EzPickle):
r"""
## Description
This environment builds on the [hopper](https://gymnasium.farama.org/environments/mujoco/hopper/) environment
by adding another set of legs making it possible for the robot to walk forward instead of
hop. Like other Mujoco environments, this environment aims to increase the number of independent state
and control variables as compared to the classic control environments. The walker is a
two-dimensional two-legged figure that consist of seven main body parts - a single torso at the top
(with the two legs splitting after the torso), two thighs in the middle below the torso, two legs
in the bottom below the thighs, and two feet attached to the legs on which the entire body rests.
The goal is to walk in the in the forward (right)
direction by applying torques on the six hinges connecting the seven body parts.
Gymnasium includes the following versions of the environment:
| Environment | Binding | Notes |
| ------------------------- | --------------- | ------------------------------------------- |
| Walker2d-v5 | `mujoco=>2.3.3` | Recommended (most features, the least bugs) |
| Walker2d-v4 | `mujoco=>2.1.3` | Maintained for reproducibility |
| Walker2d-v3 | `mujoco-py` | Maintained for reproducibility |
| Walker2d-v2 | `mujoco-py` | Maintained for reproducibility |
For more information see section "Version History".
## Action Space
```{figure} action_space_figures/walker2d.png
:name: walker2d
```
The action space is a `Box(-1, 1, (6,), float32)`. An action represents the torques applied at the hinge joints.
| Num | Action | Control Min | Control Max | Name (in corresponding XML file) | Joint | Type (Unit) |
|-----|----------------------------------------|-------------|-------------|----------------------------------|-------|--------------|
| 0 | Torque applied on the thigh rotor | -1 | 1 | thigh_joint | hinge | torque (N m) |
| 1 | Torque applied on the leg rotor | -1 | 1 | leg_joint | hinge | torque (N m) |
| 2 | Torque applied on the foot rotor | -1 | 1 | foot_joint | hinge | torque (N m) |
| 3 | Torque applied on the left thigh rotor | -1 | 1 | thigh_left_joint | hinge | torque (N m) |
| 4 | Torque applied on the left leg rotor | -1 | 1 | leg_left_joint | hinge | torque (N m) |
| 5 | Torque applied on the left foot rotor | -1 | 1 | foot_left_joint | hinge | torque (N m) |
## Observation Space
The observation Space consists of the following parts (in order):
- qpos (8 elements by default):* Position values of the robots's body parts.
- qvel (9 elements):* The velocities of these individual body parts,
(their derivatives).
By default, the observation does not include the robot's x-coordinate (`rootx`).
This can be be included by passing `exclude_current_positions_from_observation=False` during construction.
In this case, the observation space will be a `Box(-Inf, Inf, (18,), float64)`, where the first observation element is the x--coordinate of the robot.
Regardless of whether `exclude_current_positions_from_observation` is set to true or false, the x- and y-coordinates are returned in `info` with keys `"x_position"` and `"y_position"`, respectively.
By default, observation is a `Box(-Inf, Inf, (17,), float64)` where the elements correspond to the following:
| Num | Observation | Min | Max | Name (in corresponding XML file) | Joint | Type (Unit) |
| --- | -------------------------------------------------- | ---- | --- | -------------------------------- | ----- | ------------------------ |
| excluded | x-coordinate of the torso | -Inf | Inf | rootx | slide | position (m) |
| 0 | z-coordinate of the torso (height of Walker2d) | -Inf | Inf | rootz | slide | position (m) |
| 1 | angle of the torso | -Inf | Inf | rooty | hinge | angle (rad) |
| 2 | angle of the thigh joint | -Inf | Inf | thigh_joint | hinge | angle (rad) |
| 3 | angle of the leg joint | -Inf | Inf | leg_joint | hinge | angle (rad) |
| 4 | angle of the foot joint | -Inf | Inf | foot_joint | hinge | angle (rad) |
| 5 | angle of the left thigh joint | -Inf | Inf | thigh_left_joint | hinge | angle (rad) |
| 6 | angle of the left leg joint | -Inf | Inf | leg_left_joint | hinge | angle (rad) |
| 7 | angle of the left foot joint | -Inf | Inf | foot_left_joint | hinge | angle (rad) |
| 8 | velocity of the x-coordinate of the torso | -Inf | Inf | rootx | slide | velocity (m/s) |
| 9 | velocity of the z-coordinate (height) of the torso | -Inf | Inf | rootz | slide | velocity (m/s) |
| 10 | angular velocity of the angle of the torso | -Inf | Inf | rooty | hinge | angular velocity (rad/s) |
| 11 | angular velocity of the thigh hinge | -Inf | Inf | thigh_joint | hinge | angular velocity (rad/s) |
| 12 | angular velocity of the leg hinge | -Inf | Inf | leg_joint | hinge | angular velocity (rad/s) |
| 13 | angular velocity of the foot hinge | -Inf | Inf | foot_joint | hinge | angular velocity (rad/s) |
| 14 | angular velocity of the thigh hinge | -Inf | Inf | thigh_left_joint | hinge | angular velocity (rad/s) |
| 15 | angular velocity of the leg hinge | -Inf | Inf | leg_left_joint | hinge | angular velocity (rad/s) |
| 16 | angular velocity of the foot hinge | -Inf | Inf | foot_left_joint | hinge | angular velocity (rad/s) |
## Rewards
The total reward is: ***reward*** *=* *healthy_reward bonus + forward_reward - ctrl_cost*.
- *healthy_reward*:
Every timestep that the Walker2d is alive, it receives a fixed reward of value `healthy_reward`,
- *forward_reward*:
A reward for moving forward,
this reward would be positive if the Swimmer moves forward (in the positive $x$ direction / in the right direction).
$w_{forward} \times \frac{dx}{dt}$, where
$dx$ is the displacement of the (front) "tip" ($x_{after-action} - x_{before-action}$),
$dt$ is the time between actions, which depends on the `frame_skip` parameter (default is 4),
and `frametime` which is 0.002 - so the default is $dt = 4 \times 0.002 = 0.008$,
$w_{forward}$ is the `forward_reward_weight` (default is $1$).
- *ctrl_cost*:
A negative reward to penalize the Walker2d for taking actions that are too large.
$w_{control} \times \\|action\\|_2^2$,
where $w_{control}$ is `ctrl_cost_weight` (default is $10^{-3}$).
`info` contains the individual reward terms.
## Starting State
The initial position state is $[0, 1.25, 0, 0, 0, 0, 0, 0, 0] + \mathcal{U}_{[-reset\_noise\_scale \times 1_{9}, reset\_noise\_scale \times 1_{9}]}$.
The initial velocity state is $\mathcal{U}_{[-reset\_noise\_scale \times 1_{9}, reset\_noise\_scale \times 1_{9}]}$.
where $\mathcal{U}$ is the multivariate uniform continuous distribution.
Note that the z-coordinate is non-zero so that the walker2d can stand up immediately.
## Episode End
#### Termination
If `terminate_when_unhealthy is True` (which is the default), the environment terminates when the Walker2d is unhealthy.
The Walker2d is unhealthy if any of the following happens:
1. Any of the state space values is no longer finite
2. The height of the walker is ***not*** in the closed interval specified by `healthy_z_range`
3. The absolute value of the angle (`observation[1]` if `exclude_current_positions_from_observation=False`, else `observation[2]`) is ***not*** in the closed interval specified by `healthy_angle_range`
#### Truncation
The default duration of an episode is 1000 timesteps
## Arguments
Walker2d provides a range of parameters to modify the observation space, reward function, initial state, and termination condition.
These parameters can be applied during `gymnasium.make` in the following way:
```python
import gymnasium as gym
env = gym.make('Walker2d-v5', ctrl_cost_weight=1e-3, ...)
```
| Parameter | Type | Default | Description |
| -------------------------------------------- | --------- | ---------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------- |
| `xml_file` | **str** |`"walker2d_v5.xml"`| Path to a MuJoCo model |
| `forward_reward_weight` | **float** | `1` | Weight for _forward_reward_ term (see section on reward) |
| `ctrl_cost_weight` | **float** | `1e-3` | Weight for _ctr_cost_ term (see section on reward) |
| `healthy_reward` | **float** | `1` | Weight for _healthy_reward_ reward (see section on reward) |
| `terminate_when_unhealthy` | **bool** | `True` | If true, issue a done signal if the z-coordinate of the walker is no longer healthy |
| `healthy_z_range` | **tuple** | `(0.8, 2)` | The z-coordinate of the torso of the walker must be in this range to be considered healthy |
| `healthy_angle_range` | **tuple** | `(-1, 1)` | The angle must be in this range to be considered healthy |
| `reset_noise_scale` | **float** | `5e-3` | 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
* v5:
- Minimum `mujoco` version is now 2.3.3.
- Added support for fully custom/third party `mujoco` models using the `xml_file` argument (previously only a few changes could be made to the existing models).
- Added `default_camera_config` argument, a dictionary for setting the `mj_camera` properties, mainly useful for custom environments.
- Added `env.observation_structure`, a dictionary for specifying the observation space compose (e.g. `qpos`, `qvel`), useful for building tooling and wrappers for the MuJoCo environments.
- Return a non-empty `info` with `reset()`, previously an empty dictionary was returned, the new keys are the same state information as `step()`.
- Added `frame_skip` argument, used to configure the `dt` (duration of `step()`), default varies by environment check environment documentation pages.
- In v2, v3 and v4 the models have different friction values for the two feet (left foot friction == 1.9 and right foot friction == 0.9). The `Walker-v5` model is updated to have the same friction for both feet (set to 1.9). This causes the Walker2d's the right foot to slide less on the surface and therefore require more force to move (related [Github issue](https://github.com/Farama-Foundation/Gymnasium/issues/477)).
- Fixed bug: `healthy_reward` was given on every step (even if the Walker2D is unhealthy), now it is only given if the Walker2d is healthy. The `info` "reward_survive" is updated with this change (related [Github issue](https://github.com/Farama-Foundation/Gymnasium/issues/526)).
- Restored the `xml_file` argument (was removed in `v4`).
- Added individual reward terms in `info` (`info["reward_forward"]`, info`["reward_ctrl"]`, `info["reward_survive"]`).
- Added `info["z_distance_from_origin"]` which is equal to the vertical distance of the "torso" body from its initial position.
* v4: All MuJoCo environments now use the MuJoCo bindings in mujoco >= 2.1.3
* v3: Support for `gymnasium.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)
"""
metadata = {
"render_modes": [
"human",
"rgb_array",
"depth_array",
],
}
def __init__(
self,
xml_file: str = "walker2d_v5.xml",
frame_skip: int = 4,
default_camera_config: Dict[str, float] = DEFAULT_CAMERA_CONFIG,
forward_reward_weight: float = 1.0,
ctrl_cost_weight: float = 1e-3,
healthy_reward: float = 1.0,
terminate_when_unhealthy: bool = True,
healthy_z_range: Tuple[float, float] = (0.8, 2.0),
healthy_angle_range: Tuple[float, float] = (-1.0, 1.0),
reset_noise_scale: float = 5e-3,
exclude_current_positions_from_observation: bool = True,
**kwargs,
):
utils.EzPickle.__init__(
self,
xml_file,
frame_skip,
default_camera_config,
forward_reward_weight,
ctrl_cost_weight,
healthy_reward,
terminate_when_unhealthy,
healthy_z_range,
healthy_angle_range,
reset_noise_scale,
exclude_current_positions_from_observation,
**kwargs,
)
self._forward_reward_weight = forward_reward_weight
self._ctrl_cost_weight = ctrl_cost_weight
self._healthy_reward = healthy_reward
self._terminate_when_unhealthy = terminate_when_unhealthy
self._healthy_z_range = healthy_z_range
self._healthy_angle_range = healthy_angle_range
self._reset_noise_scale = reset_noise_scale
self._exclude_current_positions_from_observation = (
exclude_current_positions_from_observation
)
MujocoEnv.__init__(
self,
xml_file,
frame_skip,
observation_space=None,
default_camera_config=default_camera_config,
**kwargs,
)
self.metadata = {
"render_modes": [
"human",
"rgb_array",
"depth_array",
],
"render_fps": int(np.round(1.0 / self.dt)),
}
obs_size = (
self.data.qpos.size
+ self.data.qvel.size
- exclude_current_positions_from_observation
)
self.observation_space = Box(
low=-np.inf, high=np.inf, shape=(obs_size,), dtype=np.float64
)
self.observation_structure = {
"skipped_qpos": 1 * exclude_current_positions_from_observation,
"qpos": self.data.qpos.size
- 1 * exclude_current_positions_from_observation,
"qvel": self.data.qvel.size,
}
@property
def healthy_reward(self):
return self.is_healthy * self._healthy_reward
def control_cost(self, action):
control_cost = self._ctrl_cost_weight * np.sum(np.square(action))
return control_cost
@property
def is_healthy(self):
z, angle = self.data.qpos[1:3]
min_z, max_z = self._healthy_z_range
min_angle, max_angle = self._healthy_angle_range
healthy_z = min_z < z < max_z
healthy_angle = min_angle < angle < max_angle
is_healthy = healthy_z and healthy_angle
return is_healthy
@property
def terminated(self):
terminated = (not self.is_healthy) and self._terminate_when_unhealthy
return terminated
def _get_obs(self):
position = self.data.qpos.flat.copy()
velocity = np.clip(self.data.qvel.flat.copy(), -10, 10)
if self._exclude_current_positions_from_observation:
position = position[1:]
observation = np.concatenate((position, velocity)).ravel()
return observation
def step(self, action):
x_position_before = self.data.qpos[0]
self.do_simulation(action, self.frame_skip)
x_position_after = self.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
healthy_reward = self.healthy_reward
rewards = forward_reward + healthy_reward
costs = ctrl_cost
observation = self._get_obs()
reward = rewards - costs
terminated = self.terminated
info = {
"reward_forward": forward_reward,
"reward_ctrl": -ctrl_cost,
"reward_survive": healthy_reward,
"x_position": x_position_after,
"z_distance_from_origin": self.data.qpos[1] - self.init_qpos[1],
"x_velocity": x_velocity,
}
if self.render_mode == "human":
self.render()
return observation, reward, terminated, False, info
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.np_random.uniform(
low=noise_low, high=noise_high, size=self.model.nv
)
self.set_state(qpos, qvel)
observation = self._get_obs()
return observation
def _get_reset_info(self):
return {
"x_position": self.data.qpos[0],
"z_distance_from_origin": self.data.qpos[1] - self.init_qpos[1],
}