def run(args): # logger make_logger(args.result_dir) # make envs env = make_env(args) # make config config = make_config(args) # make planner planner = make_planner(args, config) # make model model = make_model(args, config) # make controller controller = make_controller(args, config, model) # make simulator runner = make_runner(args) # run experiment history_x, history_u, history_g = runner.run(env, controller, planner) # plot results plot_results(args, history_x, history_u, history_g=history_g) save_plot_data(args, history_x, history_u, history_g=history_g)
def __init__(self, params): self.params = params self.env = make_env(params) self.IL = ILNetwork(params) self.collect_policy = MLPolicy(self.IL, params) if self.params.env == "CartPole": self.expert_policy = iLQRPolicy() elif self.params.env == "TwoWheeledTrack": self.expert_policy = NMPCCGMRESPolicy() self.x_training_data = np.zeros((1, self.env.config['state_size']), dtype=np.float32) self.u_training_data = np.zeros((1, self.env.config['input_size']), dtype=np.float32)
def run(args): make_logger(args.result_dir) env = make_env(args) config = make_config(args) planner = make_planner(args, config) model = make_model(args, config) controller = make_controller(args, config, model) runner = make_runner(args) history_x_all, history_u_all, history_g_all, history_cost_all = [], [], [], [ ] # this is the collection list of n_sample trajectories for iter_sample in range(args.n_sample): print("Sampling {} th trajectory generated by expert policy:".format( iter_sample)) history_x, history_u, history_g, cost = runner.run( env, controller, planner) history_x_all.append(history_x) history_u_all.append(history_u) history_g_all.append(history_g) history_cost_all.append(cost) plot_results( history_x, history_u, history_g=history_g, args=args) # no need to change now, just see the plot of the last traj save_plot_data(history_x_all, history_u_all, history_g=history_g_all, cost=history_cost_all, args=args) # save lists if args.save_anim: animator = Animator(env, args=args) animator.draw(history_x, history_g)
def __init__(self, params): self.sess = tf.Session() # build net and initialize variables when instantiate this class self.env = make_env(params) self.build_net() self.sess.run(tf.global_variables_initializer())
def __init__(self, ILNet, params): self.ILNet = ILNet self.env = make_env(params)