def show_the_model(args): """Show the problem""" floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SingleGoalProblem(floor, goal) env.x0lb[:2] = [0.1, 0.1] env.x0ub[:2] = [0.9, 0.9] if args.novio: env.out_vio_step = 1 env_name = str(env) config = pu.get_train_gym_config(env_name=env_name) sim_n = args.num v_x0, v_xf, v_traj = pu.policy_rollout(env, config, sim_n, False, True) # fig, ax = pld.get3dAxis() matplotlib.use('TkAgg') plt.switch_backend('TkAgg') fig, axes = pl.subplots(sim_n) for i in range(sim_n): sim_rst = v_traj[i] x, u, dt = sim_rst['state'], sim_rst['action'], sim_rst['dt'] ax = axes[i] floor.draw(ax) circle1 = plt.Circle((0.8, 0.8), 0.05, color='g') ax.add_artist(circle1) ax.plot(*x[:, :2].T) fig.tight_layout() plt.savefig('gallery/%s-%d-cases.pdf' % (env_name, sim_n)) plt.show()
def show_the_model(args): """Show the problem""" floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SensorSingleGoalProblem(floor, goal) if args.novio: env.out_vio_step = 1 if args.harder: env.x0lb[0] = 0.1 env.x0ub[0] = 0.9 env.x0lb[1] = 0.1 env.x0ub[1] = 0.9 env_name = str(env) print('Environment is %s' % env_name) config = pu.get_train_gym_config(env_name=env_name) sim_n = args.num v_x0, v_xf, v_traj = pu.policy_rollout(env, config, sim_n, False, True) plt.switch_backend('TkAgg') fig, axes = pl.subplots(sim_n) for i in range(sim_n): ax = axes[i] sim_rst = v_traj[i] x, u, dt = sim_rst['state'], sim_rst['action'], sim_rst['dt'] floor.draw(ax) circle1 = plt.Circle((goal[0], goal[1]), 0.05, color='g') ax.add_artist(circle1) ax.plot(*x[:, :2].T) plt.savefig('gallery/%s-%d-cases.pdf' % (env_name, sim_n)) plt.show()
def perform_massive_test(args): """Perform some simulation and see the results.""" floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SingleGoalProblem(floor, goal) env.x0lb[:2] = [0.1, 0.1] env.x0ub[:2] = [0.9, 0.9] if args.novio: env.out_vio_step = 1 env_name = str(env) config = pu.get_train_gym_config(env_name=env_name) sim_n = args.num assert sim_n == 1000 v_x0, v_xf, v_flag = pu.policy_rollout(env, config, sim_n, show=False, return_success=True) mask0 = v_flag == 1 collision = np.sum(v_flag == -1) print('succeed in %d / %d' % (np.sum(mask0), sim_n)) print('collision in %d / %d' % (collision, sim_n)) with open('succeed_log.txt', 'wt') as f: f.write('Env:%s' % env_name) f.write('Succeed: %d' % np.sum(mask0)) f.write('Collision: %d' % collision)
def construct_env(): """Return the environment""" floor = construct_default_floor_plan() glb = np.array([0.8, 0.3, 0, 0]) gub = np.array([0.8, 0.7, 0, 0]) env = RangeGoalProblem(floor, glb, gub) env.x0lb[0] = 0.1 # xmin env.x0ub[0] = 0.9 # xmax env.x0lb[1] = 0.1 # ymin env.x0ub[1] = 0.9 # ymax return env, floor
def construct_env(args): """Return the environment""" floor = construct_default_floor_plan() glb = np.array([0.1, 0.6, 0, 0]) gub = np.array([0.4, 0.9, 0, 0]) env = SensorRangeGoalProblem(floor, glb, gub) if args.novio: env.out_vio_step = 1 env.x0lb[0] = 0.1 # xmin env.x0lb[1] = 0.6 # ymin env.x0ub[0] = 0.4 # xmax env.x0ub[1] = 0.9 # ymax return env, floor
def main(): from floorPlan import construct_default_floor_plan space = construct_default_floor_plan() # create simulator sim = LidarSimulator(space, 72, 2 * np.pi, 0.5) center = np.array([0.2, 0.3]) angle = np.pi / 2 # dis = sim.observe_one(center, angle) noise = 0 obs, pnts = sim.observe_span(center, angle, noise, return_point=True) fig, ax = plt.subplots() space.draw(ax) ax.scatter(pnts[:, 0], pnts[:, 1], marker='*', color='r') ax.scatter(center[0], center[1], marker='o', color='g') fig.savefig('gallery/sensor.pdf') plt.show()
def train_the_model(args): """Train a model for this particular problem.""" # construct the fix world thing floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SingleGoalProblem(floor, goal) env.x0lb[:2] = [0.1, 0.1] env.x0ub[:2] = [0.9, 0.9] if args.novio: env.out_vio_step = 1 # env.disable_collision() # env.disable_bound_check() env_name = str(env) config = pu.get_train_gym_config(env_name=env_name, seed=np.random.randint(10000)) pu.train_a_gym_model(env, config)
def train_the_model(args): """Train a model for this particular problem.""" # construct the fix world thing floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SensorSingleGoalProblem(floor, goal) if args.novio: env.out_vio_step = 1 if args.harder: env.x0lb[0] = 0.1 env.x0ub[0] = 0.9 env.x0lb[1] = 0.1 env.x0ub[1] = 0.9 env_name = str(env) print('Environment is %s' % env_name) config = pu.get_train_gym_config(env_name=env_name, cuda=True, seed=np.random.randint(10000)) pu.train_a_gym_model(env, config)
def construct_env(args): """Return the environment""" floor = construct_default_floor_plan() glb = np.array([0.1, 0.6, 0, 0]) gub = np.array([0.4, 0.9, 0, 0]) env = SensorRangeGoalProblem(floor, glb, gub) if args.novio: env.out_vio_step = 1 env.x0lb[0] = 0.1 env.x0lb[1] = 0.6 env.x0ub[0] = 0.4 env.x0ub[1] = 0.9 def update_fun(env, alpha): env.glb[1] = 0.6 - 0.5 * alpha # tends to 0.1 env.gub[0] = 0.4 + 0.5 * alpha # tends to 0.9 env.x0lb[1] = env.glb[1] env.x0ub[0] = env.gub[0] return env return env, floor, update_fun
def perform_mass_test(args): """Do a massive simulation on this problem.""" floor = construct_default_floor_plan() goal = np.array([0.8, 0.8]) env = SensorSingleGoalProblem(floor, goal) if args.novio: env.out_vio_step = 1 env.x0lb[0] = 0.1 env.x0ub[0] = 0.9 env.x0lb[1] = 0.1 env.x0ub[1] = 0.9 env_name = str(env) env.out_vio_step = 1 # do this anyway print('Environment is %s' % env_name) config = pu.get_train_gym_config(env_name=env_name) sim_n = args.num v_x0, v_xf, v_flag = pu.policy_rollout(env, config, sim_n, show=False, return_success=True) succeed = np.sum(v_flag == 1) collision = np.sum(v_flag == -1) print('succeed in %d / %d' % (succeed, sim_n)) print('collision in %d / %d' % (collision, sim_n)) print('0.1:', np.sum(np.linalg.norm(v_xf - np.array([0.8, 0.8, 0, 0]), axis=1) < 0.1)) with open('succeed_log.txt', 'at') as f: f.write('Env:%s' % env_name) f.write('Succeed: %d' % succeed) f.write('Collision: %d' % collision) return fig, ax = pl.subplots() plt.switch_backend('TkAgg') floor.draw(ax) ax.scatter(*v_x0[mask0, :2].T, label='Succeed') ax.scatter(*v_x0[~mask0, :2].T, label='Failure') ax.legend() fig.savefig('gallery/%s-massive-sim-x0.pdf' % env_name) plt.show()