ce96b03b07a4a4bdd851aa84493c616cd291aff2,rllib/examples/env/mbmpo_env.py,HopperWrapper,reward,#HopperWrapper#Any#Any#Any#,39

Before Change

        alive_bonus = 1.0
        assert obs.ndim == 2 and action.ndim == 2
        assert obs.shape == obs_next.shape and action.shape[0] == obs.shape[0]
        vel = obs_next[:, 5]
        ctrl_cost = 1e-3 * np.sum(np.square(action), axis=1)
        reward = vel + alive_bonus - ctrl_cost
        return np.minimum(np.maximum(-1000.0, reward), 1000.0)

After Change

            np.clip(obs[:, 1], -1.0, 1.0), np.clip(obs[:, 0], -1.0, 1.0))
        // Do everything in (B,) space (single theta-, action- and
        // reward values).
        a = np.clip(action, -self.max_torque, self.max_torque)[0]
        costs = self.angle_normalize(theta) ** 2 + \
            0.1 * obs[:, 2] ** 2 + 0.001 * (a ** 2)
        return -costs

Instances