import math import random from agent.Base_Agent import Base_Agent as Agent from behaviors.custom.Step.Step import Step from world.commons.Draw import Draw from stable_baselines3 import PPO from stable_baselines3.common.vec_env import SubprocVecEnv from scripts.commons.Server import Server from scripts.commons.Train_Base import Train_Base from time import sleep import os, gym import numpy as np from math_ops.Math_Ops import Math_Ops as M from behaviors.custom.Step.Step_Generator import Step_Generator ''' Objective: Learn how to run forward using step primitive ---------- - class Basic_Run: implements an OpenAI custom gym - class Train: implements algorithms to train a new model or test an existing model ''' class sprint(gym.Env): def __init__(self, ip, server_p, monitor_p, r_type, enable_draw) -> None: self.Gen_player_pos = None self.internal_target = None self.values_l = None self.values_r = None self.reset_time = None self.behavior = None self.bias_dir = None self.robot_type = r_type self.kick_ori = 0 self.terminal = False # Args: Server IP, Agent Port, Monitor Port, Uniform No., Robot Type, Team Name, Enable Log, Enable Draw self.player = Agent(ip, server_p, monitor_p, 1, self.robot_type, "Gym", True, enable_draw) self.step_counter = 0 # to limit episode size self.ik = self.player.inv_kinematics self.target = None # Step behavior defaults self.STEP_DUR = 10 self.STEP_Z_SPAN = 0.2 self.STEP_Z_MAX = 0.7 nao_specs = self.ik.NAO_SPECS self.leg_length = nao_specs[1] + nao_specs[3] # upper leg height + lower leg height feet_y_dev = nao_specs[0] * 2 # wider step sample_time = self.player.world.robot.STEPTIME max_ankle_z = nao_specs[5] * 1.8 self.step_generator = Step_Generator(feet_y_dev, sample_time, max_ankle_z) self.DEFAULT_ARMS = np.array([-90, -90, 8, 8, 90, 90, 70, 70], np.float32) self.walk_rel_orientation = None self.walk_rel_target = None self.walk_distance = None self.act = np.zeros(16, np.float32) # memory variable # State space obs_size = 63 self.obs = np.zeros(obs_size, np.float32) self.observation_space = gym.spaces.Box(low=np.full(obs_size, -np.inf, np.float32), high=np.full(obs_size, np.inf, np.float32), dtype=np.float32) # Action space MAX = np.finfo(np.float32).max self.no_of_actions = act_size = 16 self.action_space = gym.spaces.Box(low=np.full(act_size, -MAX, np.float32), high=np.full(act_size, MAX, np.float32), dtype=np.float32) # Place ball far away to keep landmarks in FoV (head follows ball while using Step behavior) self.player.scom.unofficial_move_ball((14, 0, 0.042)) self.ball_pos = np.array([0, 0, 0]) self.player.scom.unofficial_set_play_mode("PlayOn") self.player.scom.unofficial_move_ball((0, 0, 0)) self.gait: Step_Generator = self.player.behavior.get_custom_behavior_object("Walk").env.step_generator def observe(self, init=False): r = self.player.world.robot if init: # reset variables self.act = np.zeros(16, np.float32) # memory variable self.step_counter = 0 # index observation naive normalization self.obs[0] = min(self.step_counter, 15 * 8) / 100 # simple counter: 0,1,2,3... self.obs[1] = r.loc_head_z * 3 # z coordinate (torso) self.obs[2] = r.loc_head_z_vel / 2 # z velocity (torso) self.obs[3] = r.imu_torso_roll / 15 # absolute torso roll in deg self.obs[4] = r.imu_torso_pitch / 15 # absolute torso pitch in deg self.obs[5:8] = r.gyro / 100 # gyroscope self.obs[8:11] = r.acc / 10 # accelerometer self.obs[11:17] = r.frp.get('lf', np.zeros(6)) * (10, 10, 10, 0.01, 0.01, 0.01) # left foot: relative point of origin (p) and force vector (f) -> (px,py,pz,fx,fy,fz)* self.obs[17:23] = r.frp.get('rf', np.zeros(6)) * (10, 10, 10, 0.01, 0.01, 0.01) # right foot: relative point of origin (p) and force vector (f) -> (px,py,pz,fx,fy,fz)* # *if foot is not touching the ground, then (px=0,py=0,pz=0,fx=0,fy=0,fz=0) # Joints: Forward kinematics for ankles + feet rotation + arms (pitch + roll) rel_lankle = self.player.inv_kinematics.get_body_part_pos_relative_to_hip( "lankle") # ankle position relative to center of both hip joints rel_rankle = self.player.inv_kinematics.get_body_part_pos_relative_to_hip( "rankle") # ankle position relative to center of both hip joints lf = r.head_to_body_part_transform("torso", r.body_parts['lfoot'].transform) # foot transform relative to torso rf = r.head_to_body_part_transform("torso", r.body_parts['rfoot'].transform) # foot transform relative to torso lf_rot_rel_torso = np.array( [lf.get_roll_deg(), lf.get_pitch_deg(), lf.get_yaw_deg()]) # foot rotation relative to torso rf_rot_rel_torso = np.array( [rf.get_roll_deg(), rf.get_pitch_deg(), rf.get_yaw_deg()]) # foot rotation relative to torso # pose self.obs[23:26] = rel_lankle * (8, 8, 5) self.obs[26:29] = rel_rankle * (8, 8, 5) self.obs[29:32] = lf_rot_rel_torso / 20 self.obs[32:35] = rf_rot_rel_torso / 20 self.obs[35:39] = r.joints_position[14:18] / 100 # arms (pitch + roll) # velocity self.obs[39:55] = r.joints_target_last_speed[2:18] # predictions == last action ''' Expected observations for walking state: Time step R 0 1 2 3 4 5 6 7 0 Progress 1 0 .14 .28 .43 .57 .71 .86 1 0 Left leg active T F F F F F F F F T ''' if init: # the walking parameters refer to the last parameters in effect (after a reset, they are pointless) self.obs[55] = 1 # step progress self.obs[56] = 1 # 1 if left leg is active self.obs[57] = 0 # 1 if right leg is active else: self.obs[55] = self.step_generator.external_progress # step progress self.obs[56] = float(self.step_generator.state_is_left_active) # 1 if left leg is active self.obs[57] = float(not self.step_generator.state_is_left_active) # 1 if right leg is active ''' Create internal target with a smoother variation ''' MAX_LINEAR_DIST = 0.5 MAX_LINEAR_DIFF = 0.014 # max difference (meters) per step MAX_ROTATION_DIFF = 1.6 # max difference (degrees) per step MAX_ROTATION_DIST = 45 if init: self.internal_rel_orientation = 0 self.internal_target = np.zeros(2) previous_internal_target = np.copy(self.internal_target) # ---------------------------------------------------------------- compute internal linear target rel_raw_target_size = np.linalg.norm(self.walk_rel_target) if rel_raw_target_size == 0: rel_target = self.walk_rel_target else: rel_target = self.walk_rel_target / rel_raw_target_size * min(self.walk_distance, MAX_LINEAR_DIST) internal_diff = rel_target - self.internal_target internal_diff_size = np.linalg.norm(internal_diff) if internal_diff_size > MAX_LINEAR_DIFF: self.internal_target += internal_diff * (MAX_LINEAR_DIFF / internal_diff_size) else: self.internal_target[:] = rel_target # ---------------------------------------------------------------- compute internal rotation target internal_ori_diff = np.clip(M.normalize_deg(self.walk_rel_orientation - self.internal_rel_orientation), -MAX_ROTATION_DIFF, MAX_ROTATION_DIFF) self.internal_rel_orientation = np.clip(M.normalize_deg(self.internal_rel_orientation + internal_ori_diff), -MAX_ROTATION_DIST, MAX_ROTATION_DIST) # ----------------------------------------------------------------- observations internal_target_vel = self.internal_target - previous_internal_target self.obs[58] = self.internal_target[0] / MAX_LINEAR_DIST self.obs[59] = self.internal_target[1] / MAX_LINEAR_DIST self.obs[60] = self.internal_rel_orientation / MAX_ROTATION_DIST self.obs[61] = internal_target_vel[0] / MAX_LINEAR_DIFF self.obs[62] = internal_target_vel[0] / MAX_LINEAR_DIFF return self.obs def execute_ik(self, l_pos, l_rot, r_pos, r_rot): r = self.player.world.robot # Apply IK to each leg + Set joint targets # Left leg indices, self.values_l, error_codes = (self.ik.leg(l_pos, l_rot, True, dynamic_pose=False)) r.set_joints_target_position_direct(indices, self.values_l, harmonize=False) # Right leg indices, self.values_r, error_codes = self.ik.leg(r_pos, r_rot, False, dynamic_pose=False) r.set_joints_target_position_direct(indices, self.values_r, harmonize=False) def sync(self): ''' Run a single simulation step ''' r = self.player.world.robot self.player.scom.commit_and_send(r.get_command()) self.player.scom.receive() def reset(self): # print("reset") ''' Reset and stabilize the robot Note: for some behaviors it would be better to reduce stabilization or add noise ''' self.player.scom.unofficial_set_play_mode("PlayOn") Gen_ball_pos = [15, 0, 0] self.Gen_player_pos = (random.random() * 3 - 15, 10 - random.random() * 20, 0.5) self.ball_pos = np.array(Gen_ball_pos) self.player.scom.unofficial_move_ball((Gen_ball_pos[0], Gen_ball_pos[1], Gen_ball_pos[2])) self.step_counter = 0 self.behavior = self.player.behavior r = self.player.world.robot w = self.player.world t = w.time_local_ms self.reset_time = t self.target = np.array([3, 0]) distance = np.linalg.norm(self.target - self.Gen_player_pos[:2]) self.walk_rel_target = self.target self.walk_distance = distance self.walk_rel_orientation = M.vector_angle(self.walk_rel_target) for _ in range(25): self.player.scom.unofficial_beam(self.Gen_player_pos, 0) # beam player continuously (floating above ground) self.player.behavior.execute("Zero_Bent_Knees") self.sync() # beam player to ground self.player.scom.unofficial_beam(self.Gen_player_pos, 0) r.joints_target_speed[ 0] = 0.01 # move head to trigger physics update (rcssserver3d bug when no joint is moving) self.sync() # stabilize on ground for _ in range(7): self.player.behavior.execute("Zero_Bent_Knees") self.sync() # walk to ball while True: if w.time_local_ms - self.reset_time > 2500 and not self.gait.state_is_left_active and self.gait.state_current_ts == 2: break else: if self.player.behavior.is_ready("Get_Up"): self.player.behavior.execute_to_completion("Get_Up") reset_walk = self.behavior.previous_behavior != "Walk" # reset walk if it wasn't the previous behavior self.behavior.execute_sub_behavior("Walk", reset_walk, self.walk_rel_target, True, None, True, None) # target, is_target_abs, ori, is_ori_abs, distance self.sync() self.act = np.zeros(self.no_of_actions, np.float32) return self.observe(True) def render(self, mode='human', close=False): return def close(self): Draw.clear_all() self.player.terminate() def generate_random_target(self, position, x_range=(-15, 15), y_range=(-10, 10)): while True: x = np.random.uniform(x_range[0], x_range[1]) y = np.random.uniform(y_range[0], y_range[1]) if np.linalg.norm(np.array([x, y]) - position[:2]) >= 10: break self.target = np.array([x, y]) def step(self, action): r = (self. player.world.robot) w = self.player.world internal_dist = np.linalg.norm(self.internal_target) action_mult = 1 if internal_dist > 0.2 else (0.7 / 0.2) * internal_dist + 0.3 self.walk_rel_target = M.rotate_2d_vec( (self.target[0] - r.loc_head_position[0], self.target[1] - r.loc_head_position[1]), -r.imu_torso_orientation) self.walk_distance = np.linalg.norm(self.walk_rel_target) self.walk_rel_orientation = M.vector_angle(self.walk_rel_target) * 0.5 # exponential moving average self.act = 0.8 * self.act + 0.2 * action * action_mult * 0.7 # execute Step behavior to extract the target positions of each leg (we will override these targets) lfy, lfz, rfy, rfz = self.step_generator.get_target_positions(self.step_counter == 0, self.STEP_DUR, self.STEP_Z_SPAN, self.leg_length * self.STEP_Z_MAX) # Leg IK a = self.act l_ankle_pos = (a[0] * 0.02, max(0.01, a[1] * 0.02 + lfy), a[2] * 0.01 + lfz) # limit y to avoid self collision r_ankle_pos = (a[3] * 0.02, min(a[4] * 0.02 + rfy, -0.01), a[5] * 0.01 + rfz) # limit y to avoid self collision l_foot_rot = a[6:9] * (3, 3, 5) r_foot_rot = a[9:12] * (3, 3, 5) # Limit leg yaw/pitch l_foot_rot[2] = max(0, l_foot_rot[2] + 7) r_foot_rot[2] = min(0, r_foot_rot[2] - 7) # Arms actions arms = np.copy(self.DEFAULT_ARMS) # default arms pose arm_swing = math.sin(self.step_generator.state_current_ts / self.STEP_DUR * math.pi) * 6 inv = 1 if self.step_generator.state_is_left_active else -1 arms[0:4] += a[12:16] * 4 + (-arm_swing * inv, arm_swing * inv, 0, 0) # arms pitch+roll # Set target positions self.execute_ik(l_ankle_pos, l_foot_rot, r_ankle_pos, r_foot_rot) # legs r.set_joints_target_position_direct(slice(14, 22), arms, harmonize=False) # arms self.sync() self.step_counter += 1 obs = self.observe() dirct = abs((self.target[1] - r.loc_head_position[1]) / (self.target[0] - r.loc_head_position[0])) # direction_error = abs(r.imu_torso_orientation - M.vector_angle(self.target - r.loc_head_position[:2])) reward = (r.loc_torso_velocity[0] * 0.1 + dirct * r.loc_torso_velocity[1] * 0.1 + 0.1 * math.exp(-10*abs(self.walk_rel_orientation))) if np.linalg.norm(self.target - r.loc_head_position[:2]) < 0.3: reward += 50 self.generate_random_target(self.target) if self.player.behavior.is_ready("Get_Up"): self.terminal = True elif w.time_local_ms - self.reset_time > 25000 * 2: self.terminal = True else: self.terminal = False return obs, reward, self.terminal, {} class Train(Train_Base): def __init__(self, script) -> None: super().__init__(script) def train(self, args): # --------------------------------------- Learning parameters n_envs = min(12, os.cpu_count()) n_steps_per_env = 1024 # RolloutBuffer is of size (n_steps_per_env * n_envs) minibatch_size = 64 # should be a factor of (n_steps_per_env * n_envs) total_steps = 50000000 learning_rate = 3e-4 folder_name = f'Sprint_R{self.robot_type}' model_path = f'./scripts/gyms/logs/{folder_name}/' # print("Model path:", model_path) # --------------------------------------- Run algorithm def init_env(i_env): def thunk(): return sprint(self.ip, self.server_p + i_env, self.monitor_p_1000 + i_env, self.robot_type, False) return thunk servers = Server(self.server_p, self.monitor_p_1000, n_envs + 1) # include 1 extra server for testing env = SubprocVecEnv([init_env(i) for i in range(n_envs)]) eval_env = SubprocVecEnv([init_env(n_envs)]) try: if "model_file" in args: # retrain model = PPO.load(args["model_file"], env=env, device="cuda", n_envs=n_envs, n_steps=n_steps_per_env, batch_size=minibatch_size, learning_rate=learning_rate) else: # train new model model = PPO("MlpPolicy", env=env, verbose=1, n_steps=n_steps_per_env, batch_size=minibatch_size, learning_rate=learning_rate, device="cuda") model_path = self.learn_model(model, total_steps, model_path, eval_env=eval_env, eval_freq=n_steps_per_env * 20, save_freq=n_steps_per_env * 200, backup_env_file=__file__) except KeyboardInterrupt: sleep(1) # wait for child processes print("\nctrl+c pressed, aborting...\n") servers.kill() return env.close() eval_env.close() servers.kill() def test(self, args): # Uses different server and monitor ports server = Server(self.server_p - 1, self.monitor_p, 1) env = sprint(self.ip, self.server_p - 1, self.monitor_p, self.robot_type, True) model = PPO.load(args["model_file"], env=env) try: self.export_model(args["model_file"], args["model_file"] + ".pkl", False) # Export to pkl to create custom behavior self.test_model(model, env, log_path=args["folder_dir"], model_path=args["folder_dir"]) except KeyboardInterrupt: print() env.close() server.kill()