use mp api

This commit is contained in:
Maximilian Huettenrauch 2021-06-23 18:23:37 +02:00
parent af8e868309
commit c4a698b1bc
8 changed files with 288 additions and 27 deletions

View File

@ -88,6 +88,16 @@ register(
id='ALRBallInACupSimple-v0',
entry_point='alr_envs.mujoco:ALRBallInACupEnv',
max_episode_steps=4000,
kwargs={
"simplified": True,
"reward_type": "no_context",
}
)
register(
id='ALRBallInACupPDSimple-v0',
entry_point='alr_envs.mujoco:ALRBallInACupPDEnv',
max_episode_steps=4000,
kwargs={
"simplified": True,
"reward_type": "no_context"
@ -368,6 +378,26 @@ register(
}
)
register(
id='ALRBallInACupPDSimpleDetPMP-v0',
entry_point='alr_envs.mujoco.ball_in_a_cup.biac_pd:make_detpmp_env',
kwargs={
"name": "alr_envs:ALRBallInACupPDSimple-v0",
"num_dof": 3,
"num_basis": 5,
"duration": 3.5,
"post_traj_time": 4.5,
"width": 0.0035,
# "off": -0.05,
"policy_type": "motor",
"weights_scale": 0.2,
"zero_start": True,
"zero_goal": True,
"p_gains": np.array([4./3., 2.4, 2.5, 5./3., 2., 2., 1.25]),
"d_gains": np.array([0.0466, 0.12, 0.125, 0.04166, 0.06, 0.06, 0.025])
}
)
register(
id='ALRBallInACupDetPMP-v0',
entry_point='alr_envs.utils.make_env_helpers:make_detpmp_env',

View File

@ -7,10 +7,10 @@ from gym.utils import seeding
from matplotlib import patches
from alr_envs.classic_control.utils import check_self_collision
from alr_envs.utils.mps.alr_env import AlrEnv
from mp_env_api.envs.mp_env import MpEnv
class HoleReacherEnv(AlrEnv):
class HoleReacherEnv(MpEnv):
def __init__(self, n_links: int, hole_x: Union[None, float] = None, hole_depth: Union[None, float] = None,
hole_width: float = 1., random_start: bool = False, allow_self_collision: bool = False,
@ -44,7 +44,7 @@ class HoleReacherEnv(AlrEnv):
self._start_pos = np.hstack([[np.pi / 2], np.zeros(self.n_links - 1)])
self._start_vel = np.zeros(self.n_links)
self.dt = 0.01
self._dt = 0.01
action_bound = np.pi * np.ones((self.n_links,))
state_bound = np.hstack([
@ -66,6 +66,10 @@ class HoleReacherEnv(AlrEnv):
self._steps = 0
self.seed()
@property
def dt(self) -> Union[float, int]:
return self._dt
def step(self, action: np.ndarray):
"""
A single step with an action in joint velocity space

View File

@ -5,10 +5,10 @@ import numpy as np
from gym import spaces
from gym.utils import seeding
from alr_envs.utils.mps.alr_env import AlrEnv
from mp_env_api.envs.mp_env import MpEnv
class SimpleReacherEnv(AlrEnv):
class SimpleReacherEnv(MpEnv):
"""
Simple Reaching Task without any physics simulation.
Returns no reward until 150 time steps. This allows the agent to explore the space, but requires precise actions

View File

@ -6,10 +6,10 @@ import numpy as np
from gym.utils import seeding
from alr_envs.classic_control.utils import check_self_collision
from alr_envs.utils.mps.alr_env import AlrEnv
from mp_env_api.envs.mp_env import MpEnv
class ViaPointReacher(AlrEnv):
class ViaPointReacher(MpEnv):
def __init__(self, n_links, random_start: bool = True, via_target: Union[None, Iterable] = None,
target: Union[None, Iterable] = None, allow_self_collision=False, collision_penalty=1000):

View File

@ -1,3 +1,4 @@
from alr_envs.mujoco.reacher.alr_reacher import ALRReacherEnv
from alr_envs.mujoco.balancing import BalancingEnv
from alr_envs.mujoco.ball_in_a_cup.ball_in_a_cup import ALRBallInACupEnv
# from alr_envs.mujoco.ball_in_a_cup.ball_in_a_cup import ALRBallInACupEnv
from alr_envs.mujoco.ball_in_a_cup.biac_pd import ALRBallInACupPDEnv

View File

@ -3,9 +3,8 @@ from alr_envs.mujoco import alr_reward_fct
class BallInACupReward(alr_reward_fct.AlrReward):
def __init__(self, sim_time):
self.sim_time = sim_time
def __init__(self, env):
self.env = env
self.collision_objects = ["cup_geom1", "cup_geom2", "cup_base_contact_below",
"wrist_palm_link_convex_geom",
"wrist_pitch_link_convex_decomposition_p1_geom",
@ -32,7 +31,8 @@ class BallInACupReward(alr_reward_fct.AlrReward):
self.reset(None)
def reset(self, context):
self.ball_traj = np.zeros(shape=(self.sim_time, 3))
# self.sim_time = self.env.sim.dtsim_time
self.ball_traj = [] # np.zeros(shape=(self.sim_time, 3))
self.dists = []
self.dists_final = []
self.costs = []
@ -40,23 +40,24 @@ class BallInACupReward(alr_reward_fct.AlrReward):
self.angle_costs = []
self.cup_angles = []
def compute_reward(self, action, env):
self.ball_id = env.sim.model._body_name2id["ball"]
self.ball_collision_id = env.sim.model._geom_name2id["ball_geom"]
self.goal_id = env.sim.model._site_name2id["cup_goal"]
self.goal_final_id = env.sim.model._site_name2id["cup_goal_final"]
self.collision_ids = [env.sim.model._geom_name2id[name] for name in self.collision_objects]
def compute_reward(self, action):
self.ball_id = self.env.sim.model._body_name2id["ball"]
self.ball_collision_id = self.env.sim.model._geom_name2id["ball_geom"]
self.goal_id = self.env.sim.model._site_name2id["cup_goal"]
self.goal_final_id = self.env.sim.model._site_name2id["cup_goal_final"]
self.collision_ids = [self.env.sim.model._geom_name2id[name] for name in self.collision_objects]
ball_in_cup = self.check_ball_in_cup(env.sim, self.ball_collision_id)
ball_in_cup = self.check_ball_in_cup(self.env.sim, self.ball_collision_id)
# Compute the current distance from the ball to the inner part of the cup
goal_pos = env.sim.data.site_xpos[self.goal_id]
ball_pos = env.sim.data.body_xpos[self.ball_id]
goal_final_pos = env.sim.data.site_xpos[self.goal_final_id]
goal_pos = self.env.sim.data.site_xpos[self.goal_id]
ball_pos = self.env.sim.data.body_xpos[self.ball_id]
goal_final_pos = self.env.sim.data.site_xpos[self.goal_final_id]
self.dists.append(np.linalg.norm(goal_pos - ball_pos))
self.dists_final.append(np.linalg.norm(goal_final_pos - ball_pos))
self.ball_traj[env._steps, :] = ball_pos
cup_quat = np.copy(env.sim.data.body_xquat[env.sim.model._body_name2id["cup"]])
# self.ball_traj[self.env._steps, :] = ball_pos
self.ball_traj.append(ball_pos)
cup_quat = np.copy(self.env.sim.data.body_xquat[self.env.sim.model._body_name2id["cup"]])
cup_angle = np.arctan2(2 * (cup_quat[0] * cup_quat[1] + cup_quat[2] * cup_quat[3]),
1 - 2 * (cup_quat[1]**2 + cup_quat[2]**2))
cost_angle = (cup_angle - np.pi / 2) ** 2
@ -66,9 +67,9 @@ class BallInACupReward(alr_reward_fct.AlrReward):
action_cost = np.sum(np.square(action))
self.action_costs.append(action_cost)
self._is_collided = self.check_collision(env.sim) or env.check_traj_in_joint_limits()
self._is_collided = self.check_collision(self.env.sim) or self.env.check_traj_in_joint_limits()
if env._steps == env.sim_steps - 1 or self._is_collided:
if self.env._steps == self.env.ep_length - 1 or self._is_collided:
t_min_dist = np.argmin(self.dists)
angle_min_dist = self.cup_angles[t_min_dist]
# cost_angle = (angle_min_dist - np.pi / 2)**2

View File

@ -0,0 +1,225 @@
import gym.envs.mujoco
import mujoco_py.builder
from gym import utils, spaces
import os
import numpy as np
import gym.envs.mujoco as mujoco_env
from mp_env_api.envs.mp_env import MpEnv
from mp_env_api.envs.positional_env import PositionalEnv
from mp_env_api.mp_wrappers.detpmp_wrapper import DetPMPWrapper
from mp_env_api.utils.policies import PDControllerExtend
def make_detpmp_env(**kwargs):
name = kwargs.pop("name")
_env = gym.make(name)
policy = PDControllerExtend(_env, p_gains=kwargs.pop('p_gains'), d_gains=kwargs.pop('d_gains'))
kwargs['policy_type'] = policy
return DetPMPWrapper(_env, **kwargs)
class ALRBallInACupPDEnv(mujoco_env.MujocoEnv, PositionalEnv, MpEnv, utils.EzPickle):
def __init__(self, frame_skip=4, apply_gravity_comp=True, simplified: bool = False,
reward_type: str = None, context: np.ndarray = None):
self._steps = 0
self.xml_path = os.path.join(os.path.dirname(os.path.abspath(__file__)), "assets",
"biac_base" + ".xml")
self.max_ctrl = np.array([150., 125., 40., 60., 5., 5., 2.])
self.j_min = np.array([-2.6, -1.985, -2.8, -0.9, -4.55, -1.5707, -2.7])
self.j_max = np.array([2.6, 1.985, 2.8, 3.14159, 1.25, 1.5707, 2.7])
self.context = context
self.apply_gravity_comp = apply_gravity_comp
self.simplified = simplified
self._start_pos = np.array([0.0, 0.58760536, 0.0, 1.36004913, 0.0, -0.32072943, -1.57])
self._start_vel = np.zeros(7)
self.sim_time = 8 # seconds
self.ep_length = 4000 # based on 8 seconds with dt = 0.02 int(self.sim_time / self.dt)
if reward_type == "no_context":
from alr_envs.mujoco.ball_in_a_cup.ball_in_a_cup_reward_simple import BallInACupReward
reward_function = BallInACupReward
elif reward_type == "contextual_goal":
from alr_envs.mujoco.ball_in_a_cup.ball_in_a_cup_reward import BallInACupReward
reward_function = BallInACupReward
else:
raise ValueError("Unknown reward type: {}".format(reward_type))
self.reward_function = reward_function(self)
mujoco_env.MujocoEnv.__init__(self, self.xml_path, frame_skip)
utils.EzPickle.__init__(self)
@property
def start_pos(self):
if self.simplified:
return self._start_pos[1::2]
else:
return self._start_pos
@property
def start_vel(self):
if self.simplified:
return self._start_vel[1::2]
else:
return self._start_vel
@property
def current_pos(self):
return self.sim.data.qpos[0:7].copy()
@property
def current_vel(self):
return self.sim.data.qvel[0:7].copy()
# def _set_action_space(self):
# if self.simplified:
# bounds = self.model.actuator_ctrlrange.copy().astype(np.float32)[1::2]
# else:
# bounds = self.model.actuator_ctrlrange.copy().astype(np.float32)
# low, high = bounds.T
# self.action_space = spaces.Box(low=low, high=high, dtype=np.float32)
# return self.action_space
def reset(self):
self.reward_function.reset(None)
return super().reset()
def reset_model(self):
init_pos_all = self.init_qpos.copy()
init_pos_robot = self._start_pos
init_vel = np.zeros_like(init_pos_all)
self._steps = 0
self._q_pos = []
self._q_vel = []
start_pos = init_pos_all
start_pos[0:7] = init_pos_robot
self.set_state(start_pos, init_vel)
return self._get_obs()
def step(self, a):
reward_dist = 0.0
angular_vel = 0.0
reward_ctrl = - np.square(a).sum()
# if self.simplified:
# tmp = np.zeros(7)
# tmp[1::2] = a
# a = tmp
if self.apply_gravity_comp:
a = a + self.sim.data.qfrc_bias[:len(a)].copy() / self.model.actuator_gear[:, 0]
try:
self.do_simulation(a, self.frame_skip)
crash = False
except mujoco_py.builder.MujocoException:
crash = True
# joint_cons_viol = self.check_traj_in_joint_limits()
ob = self._get_obs()
if not crash:
reward, success, is_collided = self.reward_function.compute_reward(a)
done = success or is_collided # self._steps == self.sim_steps - 1
self._steps += 1
else:
reward = -2000
success = False
is_collided = False
done = True
return ob, reward, done, dict(reward_dist=reward_dist,
reward_ctrl=reward_ctrl,
velocity=angular_vel,
# traj=self._q_pos,
action=a,
q_pos=self.sim.data.qpos[0:7].ravel().copy(),
q_vel=self.sim.data.qvel[0:7].ravel().copy(),
is_success=success,
is_collided=is_collided, sim_crash=crash)
def check_traj_in_joint_limits(self):
return any(self.current_pos > self.j_max) or any(self.current_pos < self.j_min)
# TODO: extend observation space
def _get_obs(self):
theta = self.sim.data.qpos.flat[:7]
return np.concatenate([
np.cos(theta),
np.sin(theta),
# self.get_body_com("target"), # only return target to make problem harder
[self._steps],
])
# TODO
@property
def active_obs(self):
return np.hstack([
[False] * 7, # cos
[False] * 7, # sin
# [True] * 2, # x-y coordinates of target distance
[False] # env steps
])
# These functions are for the task with 3 joint actuations
def extend_des_pos(self, des_pos):
des_pos_full = self._start_pos.copy()
des_pos_full[1] = des_pos[0]
des_pos_full[3] = des_pos[1]
des_pos_full[5] = des_pos[2]
return des_pos_full
def extend_des_vel(self, des_vel):
des_vel_full = self._start_vel.copy()
des_vel_full[1] = des_vel[0]
des_vel_full[3] = des_vel[1]
des_vel_full[5] = des_vel[2]
return des_vel_full
def render(self, render_mode, **render_kwargs):
if render_mode == "plot_trajectory":
if self._steps == 1:
import matplotlib.pyplot as plt
# plt.ion()
self.fig, self.axs = plt.subplots(3, 1)
if self._steps <= 1750:
for ax, cp in zip(self.axs, self.current_pos[1::2]):
ax.scatter(self._steps, cp, s=2, marker=".")
# self.fig.show()
else:
super().render(render_mode, **render_kwargs)
if __name__ == "__main__":
env = ALRBallInACupPDEnv(reward_type="no_context", simplified=True)
# env = gym.make("alr_envs:ALRBallInACupPDSimpleDetPMP-v0")
# ctxt = np.array([-0.20869846, -0.66376693, 1.18088501])
# env.configure(ctxt)
env.reset()
env.render("human")
for i in range(16000):
# test with random actions
ac = 0.02 * env.action_space.sample()[0:7]
# ac = env.start_pos
# ac[0] += np.pi/2
obs, rew, d, info = env.step(ac)
env.render("human")
print(rew)
if d:
break
env.close()

View File

@ -110,6 +110,6 @@ if __name__ == '__main__':
# example_async("alr_envs:LongSimpleReacherDMP-v0", 4)
# example_async_contextual_sampler()
# env = gym.make("alr_envs:HoleReacherDetPMP-v1")
env_name = "alr_envs:ALRBallInACupSimpleDetPMP-v0"
env_name = "alr_envs:ALRBallInACupPDSimpleDetPMP-v0"
example_async_sampler(env_name)
# example_mp(env_name)