dodox/fancy_gym
1
0
Fork 0
Code Issues Pull Requests Projects Releases Wiki Activity

mp wrapper fixes

Browse Source
This commit is contained in:
Fabian
2022-07-06 09:05:35 +02:00
parent eddef33d9a
commit 6704c9d63a
43 changed files with 302 additions and 608 deletions
Download Patch File Download Diff File Expand all files Collapse all files
alr_envs/alr/mujoco/hopper_jump/__init__.py
+1
View File
@@ -1 +1,2 @@
from .mp_wrapper import MPWrapper
alr_envs/alr/mujoco/hopper_jump/hopper_jump.py
+104 -189
View File
@@ -1,3 +1,4 @@
import copy
from typing import Optional
from gym.envs.mujoco.hopper_v3 import HopperEnv
@@ -7,10 +8,11 @@ import os
MAX_EPISODE_STEPS_HOPPERJUMP = 250
class ALRHopperJumpEnv(HopperEnv):
class HopperJumpEnv(HopperEnv):
"""
Initialization changes to normal Hopper:
- terminate_when_unhealthy: True -> False
- healthy_reward: 1.0 -> 2.0
- healthy_z_range: (0.7, float('inf')) -> (0.5, float('inf'))
- healthy_angle_range: (-0.2, 0.2) -> (-float('inf'), float('inf'))
- exclude_current_positions_from_observation: True -> False
@@ -21,24 +23,28 @@ class ALRHopperJumpEnv(HopperEnv):
xml_file='hopper_jump.xml',
forward_reward_weight=1.0,
ctrl_cost_weight=1e-3,
healthy_reward=1.0,
penalty=0.0,
healthy_reward=2.0, # 1 step
contact_weight=2.0, # 0 step
height_weight=10.0, # 3 step
dist_weight=3.0, # 3 step
terminate_when_unhealthy=False,
healthy_state_range=(-100.0, 100.0),
healthy_z_range=(0.5, float('inf')),
healthy_angle_range=(-float('inf'), float('inf')),
reset_noise_scale=5e-3,
exclude_current_positions_from_observation=False,
sparse=False,
):
self._steps = 0
self.sparse = sparse
self._height_weight = height_weight
self._dist_weight = dist_weight
self._contact_weight = contact_weight
self.max_height = 0
# self.penalty = penalty
self.goal = 0
self._floor_geom_id = None
self._foot_geom_id = None
self._steps = 0
self.contact_with_floor = False
self.init_floor_contact = False
self.has_left_floor = False
@@ -49,12 +55,12 @@ class ALRHopperJumpEnv(HopperEnv):
healthy_state_range, healthy_z_range, healthy_angle_range, reset_noise_scale,
exclude_current_positions_from_observation)
# increase initial height
self.init_qpos[1] = 1.5
def step(self, action):
self._steps += 1
self._floor_geom_id = self.model.geom_name2id('floor')
self._foot_geom_id = self.model.geom_name2id('foot_geom')
self.do_simulation(action, self.frame_skip)
height_after = self.get_body_com("torso")[2]
@@ -73,18 +79,19 @@ class ALRHopperJumpEnv(HopperEnv):
ctrl_cost = self.control_cost(action)
costs = ctrl_cost
done = False
goal_dist = np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0]))
goal_dist = np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0]))
if self.contact_dist is None and self.contact_with_floor:
self.contact_dist = goal_dist
rewards = 0
if self._steps >= MAX_EPISODE_STEPS_HOPPERJUMP:
# healthy_reward = 0 if self.context else self.healthy_reward * self._steps
healthy_reward = self.healthy_reward * 2 # * self._steps
contact_dist = self.contact_dist if self.contact_dist is not None else 5
dist_reward = self._forward_reward_weight * (-3 * goal_dist + 10 * self.max_height - 2 * contact_dist)
rewards = dist_reward + healthy_reward
if not self.sparse or (self.sparse and self._steps >= MAX_EPISODE_STEPS_HOPPERJUMP):
healthy_reward = self.healthy_reward
distance_reward = goal_dist * self._dist_weight
height_reward = (self.max_height if self.sparse else self.get_body_com("torso")[2]) * self._height_weight
contact_reward = (self.contact_dist or 5) * self._contact_weight
# dist_reward = self._forward_reward_weight * (-3 * goal_dist + 10 * self.max_height - 2 * contact_dist)
rewards = self._forward_reward_weight * (distance_reward + height_reward + contact_reward + healthy_reward)
observation = self._get_obs()
reward = rewards - costs
@@ -97,24 +104,40 @@ class ALRHopperJumpEnv(HopperEnv):
height_rew=self.max_height,
healthy_reward=self.healthy_reward * 2,
healthy=self.is_healthy,
contact_dist=self.contact_dist if self.contact_dist is not None else 0
contact_dist=self.contact_dist or 0
)
return observation, reward, done, info
def _get_obs(self):
return np.append(super()._get_obs(), self.goal)
goal_dist = self.data.get_site_xpos('foot_site') - np.array([self.goal, 0, 0])
return np.concatenate((super(HopperJumpEnv, self)._get_obs(), goal_dist.copy(), self.goal.copy()))
def reset(self, *, seed: Optional[int] = None, return_info: bool = False, options: Optional[dict] = None, ):
self.goal = self.np_random.uniform(1.4, 2.16, 1)[0] # 1.3 2.3
def reset_model(self):
super(HopperJumpEnv, self).reset_model()
self.goal = self.np_random.uniform(0.3, 1.35, 1)[0]
self.sim.model.body_pos[self.sim.model.body_name2id('goal_site_body')] = np.array([self.goal, 0, 0])
self.max_height = 0
self._steps = 0
return super().reset()
# overwrite reset_model to make it deterministic
def reset_model(self):
noise_low = -np.zeros(self.model.nq)
noise_low[3] = -0.5
noise_low[4] = -0.2
noise_low[5] = 0
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)
noise_high = np.zeros(self.model.nq)
noise_high[3] = 0
noise_high[4] = 0
noise_high[5] = 0.785
qpos = (
self.np_random.uniform(low=noise_low, high=noise_high, size=self.model.nq) +
self.init_qpos
)
qvel = (
# self.np_random.uniform(low=noise_low, high=noise_high, size=self.model.nv) +
self.init_qvel
)
self.set_state(qpos, qvel)
@@ -123,6 +146,7 @@ class ALRHopperJumpEnv(HopperEnv):
self.contact_with_floor = False
self.init_floor_contact = False
self.contact_dist = None
return observation
def _is_floor_foot_contact(self):
@@ -137,184 +161,75 @@ class ALRHopperJumpEnv(HopperEnv):
return False
class ALRHopperXYJumpEnv(ALRHopperJumpEnv):
class HopperJumpStepEnv(HopperJumpEnv):
def __init__(self,
xml_file='hopper_jump.xml',
forward_reward_weight=1.0,
ctrl_cost_weight=1e-3,
healthy_reward=1.0,
height_weight=3,
dist_weight=3,
terminate_when_unhealthy=False,
healthy_state_range=(-100.0, 100.0),
healthy_z_range=(0.5, float('inf')),
healthy_angle_range=(-float('inf'), float('inf')),
reset_noise_scale=5e-3,
exclude_current_positions_from_observation=False
):
self._height_weight = height_weight
self._dist_weight = dist_weight
super().__init__(xml_file, forward_reward_weight, ctrl_cost_weight, healthy_reward, terminate_when_unhealthy,
healthy_state_range, healthy_z_range, healthy_angle_range, reset_noise_scale,
exclude_current_positions_from_observation)
def step(self, action):
self._floor_geom_id = self.model.geom_name2id('floor')
self._foot_geom_id = self.model.geom_name2id('foot_geom')
self._steps += 1
self.do_simulation(action, self.frame_skip)
height_after = self.get_body_com("torso")[2]
site_pos_after = self.sim.data.site_xpos[self.model.site_name2id('foot_site')].copy()
site_pos_after = self.data.get_site_xpos('foot_site')
self.max_height = max(height_after, self.max_height)
# floor_contact = self._contact_checker(self._floor_geom_id, self._foot_geom_id) if not self.contact_with_floor else False
# self.init_floor_contact = floor_contact if not self.init_floor_contact else self.init_floor_contact
# self.has_left_floor = not floor_contact if self.init_floor_contact and not self.has_left_floor else self.has_left_floor
# self.contact_with_floor = floor_contact if not self.contact_with_floor and self.has_left_floor else self.contact_with_floor
floor_contact = self._is_floor_foot_contact(self._floor_geom_id,
self._foot_geom_id) if not self.contact_with_floor else False
if not self.init_floor_contact:
self.init_floor_contact = floor_contact
if self.init_floor_contact and not self.has_left_floor:
self.has_left_floor = not floor_contact
if not self.contact_with_floor and self.has_left_floor:
self.contact_with_floor = floor_contact
if self.contact_dist is None and self.contact_with_floor:
self.contact_dist = np.linalg.norm(self.sim.data.site_xpos[self.model.site_name2id('foot_site')]
- np.array([self.goal, 0, 0]))
ctrl_cost = self.control_cost(action)
healthy_reward = self.healthy_reward
height_reward = self._height_weight * height_after
goal_dist = np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0]))
goal_dist_reward = -self._dist_weight * goal_dist
dist_reward = self._forward_reward_weight * (goal_dist_reward + height_reward)
rewards = dist_reward + healthy_reward
costs = ctrl_cost
done = False
goal_dist = np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0]))
rewards = 0
if self._steps >= self.max_episode_steps:
# healthy_reward = 0 if self.context else self.healthy_reward * self._steps
healthy_reward = self.healthy_reward * 2 # * self._steps
contact_dist = self.contact_dist if self.contact_dist is not None else 5
dist_reward = self._forward_reward_weight * (-3 * goal_dist + 10 * self.max_height - 2 * contact_dist)
rewards = dist_reward + healthy_reward
# This is only for logging the distance to goal when first having the contact
has_floor_contact = self._is_floor_foot_contact() if not self.contact_with_floor else False
if not self.init_floor_contact:
self.init_floor_contact = has_floor_contact
if self.init_floor_contact and not self.has_left_floor:
self.has_left_floor = not has_floor_contact
if not self.contact_with_floor and self.has_left_floor:
self.contact_with_floor = has_floor_contact
if self.contact_dist is None and self.contact_with_floor:
self.contact_dist = goal_dist
##############################################################
observation = self._get_obs()
reward = rewards - costs
info = {
'height': height_after,
'x_pos': site_pos_after,
'max_height': self.max_height,
'goal': self.goal,
'max_height': copy.copy(self.max_height),
'goal': copy.copy(self.goal),
'goal_dist': goal_dist,
'height_rew': self.max_height,
'healthy_reward': self.healthy_reward * 2,
'healthy': self.is_healthy,
'contact_dist': self.contact_dist if self.contact_dist is not None else 0
}
return observation, reward, done, info
def reset_model(self):
self.init_qpos[1] = 1.5
self._floor_geom_id = self.model.geom_name2id('floor')
self._foot_geom_id = self.model.geom_name2id('foot_geom')
noise_low = -np.zeros(self.model.nq)
noise_low[3] = -0.5
noise_low[4] = -0.2
noise_low[5] = 0
noise_high = np.zeros(self.model.nq)
noise_high[3] = 0
noise_high[4] = 0
noise_high[5] = 0.785
rnd_vec = self.np_random.uniform(low=noise_low, high=noise_high, size=self.model.nq)
qpos = self.init_qpos + rnd_vec
qvel = self.init_qvel
self.set_state(qpos, qvel)
observation = self._get_obs()
self.has_left_floor = False
self.contact_with_floor = False
self.init_floor_contact = False
self.contact_dist = None
return observation
def reset(self):
super().reset()
self.goal = self.np_random.uniform(0.3, 1.35, 1)[0]
self.sim.model.body_pos[self.sim.model.body_name2id('goal_site_body')] = np.array([self.goal, 0, 0])
return self.reset_model()
def _get_obs(self):
goal_diff = self.sim.data.site_xpos[self.model.site_name2id('foot_site')].copy() \
- np.array([self.goal, 0, 0])
return np.concatenate((super(ALRHopperXYJumpEnv, self)._get_obs(), goal_diff))
def set_context(self, context):
# context is 4 dimensional
qpos = self.init_qpos
qvel = self.init_qvel
qpos[-3:] = context[:3]
self.goal = context[-1]
self.set_state(qpos, qvel)
self.sim.model.body_pos[self.sim.model.body_name2id('goal_site_body')] = np.array([self.goal, 0, 0])
return self._get_obs()
class ALRHopperXYJumpEnvStepBased(ALRHopperXYJumpEnv):
def __init__(
self,
xml_file='hopper_jump.xml',
forward_reward_weight=1.0,
ctrl_cost_weight=1e-3,
healthy_reward=0.0,
penalty=0.0,
context=True,
terminate_when_unhealthy=False,
healthy_state_range=(-100.0, 100.0),
healthy_z_range=(0.5, float('inf')),
healthy_angle_range=(-float('inf'), float('inf')),
reset_noise_scale=5e-3,
exclude_current_positions_from_observation=False,
max_episode_steps=250,
height_scale=10,
dist_scale=3,
healthy_scale=2
):
self.height_scale = height_scale
self.dist_scale = dist_scale
self.healthy_scale = healthy_scale
super().__init__(xml_file, forward_reward_weight, ctrl_cost_weight, healthy_reward, penalty, context,
terminate_when_unhealthy, healthy_state_range, healthy_z_range, healthy_angle_range,
reset_noise_scale, exclude_current_positions_from_observation, max_episode_steps)
def step(self, action):
self._floor_geom_id = self.model.geom_name2id('floor')
self._foot_geom_id = self.model.geom_name2id('foot_geom')
self._steps += 1
self.do_simulation(action, self.frame_skip)
height_after = self.get_body_com("torso")[2]
site_pos_after = self.sim.data.site_xpos[self.model.site_name2id('foot_site')].copy()
self.max_height = max(height_after, self.max_height)
ctrl_cost = self.control_cost(action)
healthy_reward = self.healthy_reward * self.healthy_scale
height_reward = self.height_scale * height_after
goal_dist = np.atleast_1d(np.linalg.norm(site_pos_after - np.array([self.goal, 0, 0], dtype=object)))[0]
goal_dist_reward = -self.dist_scale * goal_dist
dist_reward = self._forward_reward_weight * (goal_dist_reward + height_reward)
reward = -ctrl_cost + healthy_reward + dist_reward
done = False
observation = self._get_obs()
###########################################################
# This is only for logging the distance to goal when first having the contact
##########################################################
floor_contact = self._is_floor_foot_contact(self._floor_geom_id,
self._foot_geom_id) if not self.contact_with_floor else False
if not self.init_floor_contact:
self.init_floor_contact = floor_contact
if self.init_floor_contact and not self.has_left_floor:
self.has_left_floor = not floor_contact
if not self.contact_with_floor and self.has_left_floor:
self.contact_with_floor = floor_contact
if self.contact_dist is None and self.contact_with_floor:
self.contact_dist = np.linalg.norm(self.sim.data.site_xpos[self.model.site_name2id('foot_site')]
- np.array([self.goal, 0, 0]))
info = {
'height': height_after,
'x_pos': site_pos_after,
'max_height': self.max_height,
'goal': self.goal,
'goal_dist': goal_dist,
'height_rew': self.max_height,
'healthy_reward': self.healthy_reward * self.healthy_reward,
'healthy': self.is_healthy,
'contact_dist': self.contact_dist if self.contact_dist is not None else 0
'height_rew': height_reward,
'healthy_reward': healthy_reward,
'healthy': copy.copy(self.is_healthy),
'contact_dist': copy.copy(self.contact_dist) or 0
}
return observation, reward, done, info
alr_envs/alr/mujoco/hopper_jump/mp_wrapper.py
+1
View File
@@ -8,6 +8,7 @@ from alr_envs.black_box.raw_interface_wrapper import RawInterfaceWrapper
class MPWrapper(RawInterfaceWrapper):
# Random x goal + random init pos
@property
def context_mask(self):
return np.hstack([
[False] * (2 + int(not self.exclude_current_positions_from_observation)), # position