Spaces:
Runtime error
Runtime error
import numpy as np | |
import os | |
import pybullet as p | |
import random | |
from cliport.tasks import primitives | |
from cliport.tasks.grippers import Spatula | |
from cliport.tasks.task import Task | |
from cliport.utils import utils | |
import numpy as np | |
from cliport.tasks.task import Task | |
from cliport.utils import utils | |
import pybullet as p | |
class BallInBowlObstacleCourse(Task): | |
"""Navigate through a maze of blocks, pick up balls of different colors and place them in the corresponding colored bowls.""" | |
def __init__(self): | |
super().__init__() | |
self.max_steps = 20 | |
self.lang_template = "navigate through the maze and place the {color} ball in the {color} bowl" | |
self.task_completed_desc = "done placing balls in bowls." | |
self.additional_reset() | |
def reset(self, env): | |
super().reset(env) | |
# Add blocks to form a maze. | |
block_size = (0.04, 0.04, 0.04) | |
block_urdf = 'block/small.urdf' | |
for _ in range(10): | |
block_pose = self.get_random_pose(env, block_size) | |
env.add_object(block_urdf, block_pose, category='fixed') | |
# Add balls of different colors. | |
ball_size = (0.04, 0.04, 0.04) | |
ball_urdf = 'ball/ball-template.urdf' | |
colors = ['red', 'blue', 'green', 'yellow'] | |
balls = [] | |
for color in colors: | |
ball_pose = self.get_random_pose(env, ball_size) | |
ball_id = env.add_object(ball_urdf, ball_pose, color=color) | |
balls.append(ball_id) | |
# Add bowls of different colors at different corners of the maze. | |
bowl_size = (0.12, 0.12, 0) | |
bowl_urdf = 'bowl/bowl.urdf' | |
bowls = [] | |
for color in colors: | |
bowl_pose = self.get_random_pose(env, bowl_size) | |
bowl_id = env.add_object(bowl_urdf, bowl_pose, color=color, category='fixed') | |
bowls.append(bowl_id) | |
# Goal: each ball is in the bowl of the same color. | |
for i in range(len(balls)): | |
self.add_goal(objs=[balls[i]], matches=np.ones((1, 1)), targ_poses=[p.getBasePositionAndOrientation(bowls[i])], replace=False, | |
rotations=True, metric='pose', params=None, step_max_reward=1/len(balls)) | |
self.lang_goals.append(self.lang_template.format(color=colors[i])) |