import numpy as np from gym import utils from gym.envs.mujoco import MuJocoPyEnv from gym.spaces import Box class ReacherEnv(MuJocoPyEnv, utils.EzPickle): metadata = { "render_modes": [ "human", "rgb_array", "depth_array", ], "render_fps": 50, } def __init__(self, **kwargs): utils.EzPickle.__init__(self, **kwargs) observation_space = Box(low=-np.inf, high=np.inf, shape=(11,), dtype=np.float64) MuJocoPyEnv.__init__( self, "reacher.xml", 2, observation_space=observation_space, **kwargs ) def step(self, a): vec = self.get_body_com("fingertip") - self.get_body_com("target") reward_dist = -np.linalg.norm(vec) reward_ctrl = -np.square(a).sum() reward = reward_dist + reward_ctrl self.do_simulation(a, self.frame_skip) if self.render_mode == "human": self.render() ob = self._get_obs() return ( ob, reward, False, False, dict(reward_dist=reward_dist, reward_ctrl=reward_ctrl), ) def viewer_setup(self): assert self.viewer is not None self.viewer.cam.trackbodyid = 0 def reset_model(self): qpos = ( self.np_random.uniform(low=-0.1, high=0.1, size=self.model.nq) + self.init_qpos ) while True: self.goal = self.np_random.uniform(low=-0.2, high=0.2, size=2) if np.linalg.norm(self.goal) < 0.2: break qpos[-2:] = self.goal qvel = self.init_qvel + self.np_random.uniform( low=-0.005, high=0.005, size=self.model.nv ) qvel[-2:] = 0 self.set_state(qpos, qvel) return self._get_obs() def _get_obs(self): theta = self.sim.data.qpos.flat[:2] return np.concatenate( [ np.cos(theta), np.sin(theta), self.sim.data.qpos.flat[2:], self.sim.data.qvel.flat[:2], self.get_body_com("fingertip") - self.get_body_com("target"), ] )