问题描述
我正在尝试使用pybullet和openAI体育馆在板上模拟球的平衡，但是现在我正在处理任何形状的东西。我的体育馆已经停放了，但是我不确定如何用pybullet来解决这个问题。这是我为一个对象编写的代码示例，我想进行移动并平衡另一个对象。这是openAI env的代码，以及到目前为止我使用pybullet渲染它的工作：
import os import gym import numpy as np import pybullet as pb import pybullet_data import math import random
class BeamEnv(gym.Env):
def __init__(self,obs_low_bounds = np.array([ 0,"TBD",-45]),obs_high_bounds = np.array([12,12,45])):
self.physicsClient = pb.connect(pb.GUI) pb.setAdditionalSearchPath(pybullet_data.getDataPath()) self._seed()
"""Environment for a ball and beam system where agent has control of tilt.
Args:
obs_low_bounds (list,optional): [target location(in),ball location(in),ball veLocity(in/s),beam angle(deg)]. Defaults to [ 0,-45].
obs_high_bounds (list,optional): As above so below. Defaults to [12,45].
"""
super(BeamEnv,self).__init__()
# Hyperparameters
self.ACC_GRAV = 386.22 # [in/s]
self.MOTOR_SPEED = 46.875 # 1.28[sec/60deg] converted to [deg/s]
self.TIME_STEP = 0.1 # [s]
# Observation Space
# _bounds = []
self.obs_low_bounds = obs_low_bounds
self.obs_high_bounds = obs_high_bounds
self._determine_max_veLocity()
self.observation_space = gym.spaces.Box(low = self.obs_low_bounds,high = self.obs_high_bounds,dtype = np.float32)
# Action Space
# increase,decrease or keep current angle
self.action_space = gym.spaces.Descrete(3)
# Reward Range
self.reward_range = (-1,1)
def _set_veLocity_bounds(self):
"""Use Inclined Plane and Kinematics Formulas
to determine min/max veLocities and set the obs_low/high_bounds
"""
# Max distance
distance_max = self.obs_high_bounds[1]
# Max Angle
ang_max = self.obs_high_bounds[3]
# Max acceletation (Inclined Plane Formula)
a_max = self.ACC_GRAV * np.sin(np.deg2rad(ang_max))
# Max VeLocity (vf^2 = v0^2 + 2ad)
vel_max = np.sqrt(2*a_max*distance_max)
# Set Bounds
self.obs_low_bounds[2] = -vel_max
self.obs_high_bounds[2] = vel_max
def _compute_observation(self): cubePos,cube0rn = pb.getBasePositionAndOrientation(self.botId) cubeEuler = pb.getEulerFromQuaternion(cubeOrn) linear,angular = pb.getBaseVeLocity(self.botId) return[cubeEuler[0],angular[0],self.vt]
def _compute_reward(self): _,cubeOrn = pb.getBasePosition(self.botId) cubeEuler = pb.getEulerFromQuaternion(cube0rn) return(1 - abs(cubeEuler[0])) *
0.1 - abs(self.vt - self.vd)
def _compute_done(self): cubePos,_ = pb.getBasePositionAndOrientation(self.botId) return cubePos[2] < 0.15 or self._envStepCounter >= 1500
def reset(self,target_location = None,ball_location = None):
pb.resetSimulation() pb.setGravity(0,-9.8) """default timestep is 1/240s""" pb.settimestep(0.01) planeId = pb.loadURDF('plane.urdf') #loading bot cubeStarPos=[0,0.001] cubeStartOrientation = pb.getQuaternionFromEuler([0,0]) path = os.path.abspath(os.path.dirname(__file__)) self.botId = pb.loadURDF(os.path.join(path,"beam.xml"),cubeStartPos,cubeStartOrientation) self._observation = self._compute_observation() return np.array(self._observation)
"""Reset the environment so the ball is not moving,there is no angle,Args:
target_location (float,optional): Desired location of ball. Defaults to random.
ball_location (float,optional): Current location of ball. Defaults to random.
Returns:
list: observation of (target location,ball location,ball veLocity,beam angle)
"""
# Set target location
if target_location is not None:
self.target_location = target_location
else:
possible_targets = list(range(self.obs_low_bounds[0],self.obs_high_bounds[0]))
self.target_location = random.choice(possible_targets)
# Set ball location
if ball_location is not None:
self.ball_location = ball_location
else:
possible_ball_locations = list(range(self.obs_low_bounds[1],self.obs_high_bounds[1]))
self.ball_location = random.choice(possible_ball_locations)
# Set Intial VeLocity and Angle to Zero
self.ball_veLocity = 0 # [in/s]
self.beam_angle = 0 # [deg]
return self._next_observation()
def _next_observation(self):
"""Determines what will happen in the next time step
Returns:
list: observation of (target location,beam angle)
"""
# Calculate acceleration (Inclined Plane Equation)
ball_acceleration = self.ACC_GRAV * np.sin(np.deg2rad(self.beam_angle))
# Calculate Next Position (x = x0 + v0t + 0.5at^2)
self.ball_location = self.ball_location + self.ball_veLocity * self.TIME_STEP + 0.5 * ball_acceleration * self.TIME_STEP**2
# Calculate New VeLocity (v = v0 + at)
self.ball_veLocity = self.ball_veLocity + ball_acceleration * self.TIME_STEP
# Clip Ball Location
self.ball_location = max(min(self.ball_location,self.obs_high_bounds[1]),self.obs_low_bounds[1])
# Clip Ball VeLocity
self.ball_veLocity = max(min(self.ball_veLocity,self.obs_high_bounds[2]),self.obs_low_bounds[2])
# Return Observation
return [self.target_location,self.ball_location,self.ball_veLocity,self.beam_angle]
def _take_action(self,action):
"""Determines change in angle due to action
Args:
action (int): increase,decrease or keep current angle
"""
# Change action to signs by subtracting by 1 ie (0,1,2) --> (-1,1)
action -= 1
# Change the angle by unit step
self.beam_angle = self.beam_angle + action * self.MOTOR_SPEED * self.TIME_STEP
# Clip
self.beam_angle = max(min(self.beam_angle,self.obs_high_bounds[3]),self.obs_low_bounds[3])
def step(self,action):
"""Take action,collect reward and get new observation
Args:
action (int): increase,decrease or keep current angle
Returns:
tuple: (observation,reward,done,info)
"""
# Take the action
self._take_action(action)
# Determine Success
if (round(abs((self.target_location - self.ball_location)),3) == 0) & (round(self.ball_veLocity,3) == 0) & (round(self.beam_angle,3) == 0):
done = True
else:
done = False
# Find Reward
reward = 1 if done else -1
# Find Next Observation
obs = self._next_observation()
# Return what happened
return obs,{}
def _seed(self,seed=None): self.np_random,seed = seeding.np_random(seed) return[seed]
这是pybullet中的一个物体，我要在其上平衡物体的板上。我不确定如何通过可变力使其移动或根据我的openAI健身脚本进行渲染。
import numpy
import pybullet as pb
physicsClient = pb.connect(pb.GUI)
> #creates plane import pybullet_data pb.setAdditionalSearchPath(pybullet_data.getDataPath()) planeId =
> pb.loadURDF('plane.urdf')
>
>
> visualShapeId = pb.createVisualShape(
> shapeType=pb.GEOM_MESH,> fileName='procedural_objects/126/126.obj',> rgbaColor=None,> meshScale=[0.1,0.1,0.1])
>
> collisionShapeId = pb.createCollisionShape(
> shapeType=pb.GEOM_MESH,0.1])
>
> #connects visual shape and collider multiBodyId = pb.createMultiBody(
> baseMass=1.0,> baseCollisionShapeIndex=collisionShapeId,> baseVisualShapeIndex=visualShapeId,> basePosition=[0,1],> baSEOrientation=pb.getQuaternionFromEuler([0,0]))
>
> pb.setGravity(0,-9.8) pb.setRealTimeSimulation(1)
> #use settimestep fn to override default time step (1/240s)
任何建议如何使pybullet生成在另一个随机形状上保持平衡的随机形状（忽略球）？
解决方法
暂无找到可以解决该程序问题的有效方法，小编努力寻找整理中！
如果你已经找到好的解决方法，欢迎将解决方案带上本链接一起发送给小编。
小编邮箱:dio#foxmail.com (将#修改为@）