def step()

in safety_gym/envs/engine.py [0:0]

• 46 lines of code
• 12 McCabe index (conditional complexity)

    def step(self, action):
        ''' Take a step and return observation, reward, done, and info '''
        action = np.array(action, copy=False)  # Cast to ndarray
        assert not self.done, 'Environment must be reset before stepping'

        info = {}

        # Set action
        action_range = self.model.actuator_ctrlrange
        # action_scale = action_range[:,1] - action_range[:, 0]
        self.data.ctrl[:] = np.clip(action, action_range[:,0], action_range[:,1]) #np.clip(action * 2 / action_scale, -1, 1)
        if self.action_noise:
            self.data.ctrl[:] += self.action_noise * self.rs.randn(self.model.nu)

        # Simulate physics forward
        exception = False
        for _ in range(self.rs.binomial(self.frameskip_binom_n, self.frameskip_binom_p)):
            try:
                self.set_mocaps()
                self.sim.step()  # Physics simulation step
            except MujocoException as me:
                print('MujocoException', me)
                exception = True
                break
        if exception:
            self.done = True
            reward = self.reward_exception
            info['cost_exception'] = 1.0
        else:
            self.sim.forward()  # Needed to get sensor readings correct!

            # Reward processing
            reward = self.reward()

            # Constraint violations
            info.update(self.cost())

            # Button timer (used to delay button resampling)
            self.buttons_timer_tick()

            # Goal processing
            if self.goal_met():
                info['goal_met'] = True
                reward += self.reward_goal
                if self.continue_goal:
                    # Update the internal layout so we can correctly resample (given objects have moved)
                    self.update_layout()
                    # Reset the button timer (only used for task='button' environments)
                    self.buttons_timer = self.buttons_resampling_delay
                    # Try to build a new goal, end if we fail
                    if self.terminate_resample_failure:
                        try:
                            self.build_goal()
                        except ResamplingError as e:
                            # Normal end of episode
                            self.done = True
                    else:
                        # Try to make a goal, which could raise a ResamplingError exception
                        self.build_goal()
                else:
                    self.done = True

        # Timeout
        self.steps += 1
        if self.steps >= self.num_steps:
            self.done = True  # Maximum number of steps in an episode reached

        return self.obs(), reward, self.done, info