replay_experiment.py

import argparse
import carb
import confuse
import cv2
import ipdb
import math
import numpy as np
import os
import rosbag
import roslaunch
import rospy
import scipy.spatial.transform as tf
import sys
import time
import traceback
import trimesh
import yaml
from omni.isaac.kit import SimulationApp
from time import sleep


def boolean_string(s):
	if s.lower() not in {'false', 'true'}:
		raise ValueError('Not a valid boolean string')
	return s.lower() == 'true'


"""
Suppose you want a stereo camera
And to have optical flow
And LiDAR (not fully supported yet) of the experiments.
This is a way in which you can re-process your info and get the results.

Suggestion: teleport is much more precise (sub mm difference). Working with velocities is fisy

This code is a bit hard-coded as it is a demonstration code.
"""
try:
	parser = argparse.ArgumentParser(description="Get Bounding Boxes")
	parser.add_argument("--experiment_folder", type=str,
	                    help="The experiment folder with the USD file and the info file")
	parser.add_argument("--headless", type=boolean_string, default=False, help="Whether run this headless or not")
	parser.add_argument("--write", type=boolean_string, default=False, help="Whether to write new cameras results")
	parser.add_argument("--write_flow", type=boolean_string, default=False, help="Whether to write optical flow")
	parser.add_argument("--write_normals", type=boolean_string, default=False, help="Whether to write normals")
	parser.add_argument("--use_teleport", type=boolean_string, default=False,
	                    help="Whether to use teleport or force joint vel, both have adv and disadv")
	parser.add_argument("--use_reindex", type=boolean_string, default=False, help="Whether to use reindexed bags")
	parser.add_argument("--bag_basename", type=str, default="7659a6c9-9fc7-4be5-bc93-5b202ff2a22b")
	parser.add_argument("--out_folder_npy", type=str, default='additional_data')
	parser.add_argument("--bag_subpath", type=str, default="")

	args, unknown = parser.parse_known_args()
	config = confuse.Configuration("NewSensor", __name__)
	config.set_args(args)

	exp_info = np.load(os.path.join(config["experiment_folder"].get(), "experiment_info.npy"), allow_pickle=True)
	exp_info = exp_info.item()

	poses_path = os.path.join(config["experiment_folder"].get(), "Viewport0", "camera")
	write_flow = config["write_flow"].get()
	write_normals = config["write_normals"].get()
	write = config["write"].get()

	CONFIG = {"display_options": 3286, "width": 1280, "height": 720, "headless": config["headless"].get()}
	kit = SimulationApp(launch_config=CONFIG, experience=f"{os.environ['EXP_PATH']}/omni.isaac.sim.python.kit")

	import utils.misc_utils
	from utils.misc_utils import *
	from utils.robot_utils import *
	from utils.simulation_utils import *
	from utils.objects_utils import *
	from utils.environment_utils import *
	from utils.human_utils import *

	simulation_environment_setup()
	rospy.init_node("new_sensor_publisher", anonymous=True, disable_signals=True, log_level=rospy.ERROR)
	local_file_prefix = "my-computer://"
	omni.usd.get_context().open_stage(local_file_prefix + config["experiment_folder"].get() + "/loaded_stage.usd", None)
	kit.update()
	kit.update()

	print("Loading stage...")
	while is_stage_loading():
		kit.update()
	print("Loading Complete")

	context = omni.usd.get_context()
	stage = context.get_stage()

	simulation_context = SimulationContext(physics_dt=1.0 / exp_info["config"]["physics_hz"].get(),
	                                       rendering_dt=1.0 / exp_info["config"]["render_hz"].get(),
	                                       stage_units_in_meters=0.01)
	simulation_context.initialize_physics()

	meters_per_unit = UsdGeom.GetStageMetersPerUnit(stage)
	set_raytracing_settings(exp_info["config"]["physics_hz"].get())
	timeline = setup_timeline(exp_info["config"])

	reversing_timeline_ratio = exp_info['reversing_timeline_ratio']
	experiment_length = exp_info['config']['experiment_length'].get()
	ratio_camera = exp_info['config']['ratio_camera'].get()
	cnt_reversal = 1
	simulation_context.stop()

	### here we add the new camera to the robot. It will be located 5 cm to the right w.r.t. the original one
	old_h_ape = []
	old_v_ape = []
	viewport_window_list = []
	ros_camera_list = []
	# omni.kit.commands.execute('CopyPrim',
	#                           path_from='/my_robot_0/camera_link/Camera',
	#                           path_to='/my_robot_0/camera_link/Camera_stereo',
	#                           exclusive_select=False)

	# set_translate(stage.GetPrimAtPath('/my_robot_0/camera_link/Camera_stereo'), [1, 0, 0])
	# component, viewport = add_camera_and_viewport("/my_robot_0/camera_link",
	#                                               exp_info["config"]["robot_sensor_size"].get(), old_h_ape, old_v_ape,
	#                                               simulation_context, 0, 0, camera_path="Camera_stereo")
	# cam_outputs = control_camera(viewport, simulation_context)
	# ros_camera_list.append([0, component, cam_outputs])
	# viewport_window_list.append(viewport)

	# omni.kit.commands.execute('CopyPrim',
	#                           path_from='/my_robot_0/camera_link/Camera_npy',
	#                           path_to='/my_robot_0/camera_link/Camera_npy_stereo',
	#                           exclusive_select=False)
	#
	# set_translate(stage.GetPrimAtPath('/my_robot_0/camera_link/Camera_npy_stereo'), [1, 0, 0])

	# viewport_npy, _ = create_viewport("/my_robot_0/camera_link/Camera_npy_stereo", config["headless"].get(),
	#                                   0, exp_info["config"]["npy_sensor_size"].get(), old_h_ape, old_v_ape, simulation_context)
	# viewport_window_list.append(viewport_npy)

	viewport_npy, _ = create_viewport("/my_robot_0/camera_link/Camera_npy", config["headless"].get(),
	                                  0, exp_info["config"]["npy_sensor_size"].get(), old_h_ape, old_v_ape, simulation_context)
	viewport_window_list.append(viewport_npy)

	is_rtx = exp_info["config"]["rtx_mode"].get()
	if is_rtx:
		set_raytracing_settings(exp_info["config"]["physics_hz"].get())
	else:
		set_pathtracing_settings(exp_info["config"]["physics_hz"].get())

	simulation_context.play()
	for _ in range(5): simulation_context.render()
	old_v_ape = [2.32] * len(old_v_ape)  # todo this is harcoded
	for index, cam in enumerate(viewport_window_list):
		simulation_context.step(render=False)
		simulation_context.render()
		camera = stage.GetPrimAtPath(cam.get_active_camera())
		camera.GetAttribute("horizontalAperture").Set(old_h_ape[index])
		camera.GetAttribute("verticalAperture").Set(old_v_ape[index])
	simulation_context.stop()

	_clock_graph = add_clock()  # add ROS clock
	og.Controller.evaluate_sync(_clock_graph)

	# add a new sensor
	lidars = []
	# sensor = add_lidar(f"/my_robot_0/yaw_link", [0, 0, -.1], [0, 0, 0], is_3d=True, is_2d=False)
	# lidars.append(sensor)
	kit.update()

	cnt_tf = -1
	use_teleport = config["use_teleport"].get()
	use_reindex = config["use_reindex"].get()
	id_bag = 0
	bag_path = os.path.join(config["experiment_folder"].get(), config['bag_subpath'].get(),
	                        f"{config['bag_basename'].get()}_{id_bag}.bag")
	joint_order = ['x_joint', 'y_joint', 'z_joint', 'roll_joint', 'pitch_joint', 'yaw_joint']
	joint_position = []
	joint_velocity = []
	joint_time = []
	robot_pose = []
	started = use_reindex
	while os.path.exists(bag_path):
		bag = rosbag.Bag(bag_path)
		for topic, msg, t in bag.read_messages(
				topics=["/my_robot_0/joint_states", "/my_robot_0/odom", "/starting_experiment"]):
			if not started:
				if topic == "/starting_experiment":
					started = True
					continue
				else:
					continue
			if 'joint' in topic:
				joint_position.append(msg.position)
				joint_velocity.append(msg.velocity)
				joint_time.append(msg.header.stamp)
			else:
				robot_pose.append([msg.pose.pose.position, msg.pose.pose.orientation])
		id_bag += 1
		bag_path = os.path.join(config["experiment_folder"].get(), config['bag_subpath'].get(),
		                        f"{config['bag_basename'].get()}_{id_bag}.bag")
	if len(joint_position) == 0:
		print("No bag found")
		sys.exit(-1)
	ratio_tf = exp_info['config']['ratio_tf'].get()
	init_x, init_y, init_z, init_roll, init_pitch, init_yaw = get_robot_joint_init_loc('/my_robot_0')
	init_pos = np.array([init_x, init_y, init_z])
	init_rot = np.array([init_roll, init_pitch, init_yaw])
	change_collision_at_path(False,paths=['/my_robot_0/camera_link/Cube.physics:collisionEnabled','/my_robot_0/yaw_link/visuals.physics:collisionEnabled'])
	kit.update()
	set_drone_joints_init_loc('/my_robot_0', [0, 0, 0], [0,0,0], 300, lower_zlim=0) # todo use actual limit from simulation
	kit.update()
	simulation_context.play()
	for _ in range(5):
		simulation_context.step(render=False)
		simulation_context.render()

	timeline.set_auto_update(False)

	timeline.set_current_time(min(- 1 / (exp_info['config']["physics_hz"].get() / ratio_camera),
	                              -abs(exp_info['config']["bootstrap_exploration"].get())))
	simulation_step = int(timeline.get_current_time() * exp_info['config']["physics_hz"].get()) - 1

	out_dir_npy = os.path.join(config['experiment_folder'].get(), config['out_folder_npy'].get())
	if write_flow:
		_tmp = extension_custom.MyRecorder()
		_tmp.on_startup()
		_settings = _tmp.get_default_settings()
		_settings["rgb"]["enabled"] = False
		_settings["motion-vector"]["enabled"] = write_flow
		_settings["motion-vector"]["colorize"] = False
		_settings["motion-vector"]["npy"] = True
		my_recorder_flow = recorder_setup(_settings, out_dir_npy, True, 0)
		my_recorder_flow._enable_record = False

	if write_normals:
		_tmp = extension_custom.MyRecorder()
		_tmp.on_startup()
		_settings = _tmp.get_default_settings()
		_settings["rgb"]["enabled"] = True
		_settings["normals"]["enabled"] = write_normals
		_settings["motion-vector"]["colorize"] = False
		_settings["motion-vector"]["npy"] = True
		my_recorder_normals = recorder_setup(_settings, out_dir_npy, True, 0)
		my_recorder_normals._enable_record = False

	if write:
		_tmp = exp_info['config']['_recorder_settings'].get()
		_tmp["depth"]["enabled"] = False
		_tmp["depthLinear"]["enabled"] = False
		_tmp["semantic"]["enabled"] = False
		_tmp["normals"]["enabled"] = False
		_tmp["bbox_2d_loose"]["enabled"] = False
		_tmp["bbox_2d_tight"]["enabled"] = False
		_tmp["bbox_3d"]["enabled"] = False

		my_recorder = recorder_setup(_tmp, out_dir_npy, True, 0)
		my_recorder._enable_record = False

	# how to hide dynamic content
	dynamicprims = []
	for prim in stage.Traverse():
		if 'my_human' in str(prim.GetPath()).lower():
			dynamicprims.append(prim)
	for prim in stage.GetPrimAtPath("/World").GetChildren()[6:]:
		dynamicprims.append(prim)
	toggle_dynamic_objects(dynamicprims, False)

	forward = True
	while kit.is_running():
		simulation_step += 1
		if simulation_step == 0:
			_dc = dynamic_control_interface()
			handle = _dc.get_rigid_body('/my_robot_0/yaw_link')
			if not use_teleport:
				art = _dc.get_articulation('/my_robot_0')
				joints = []
				_dc.wake_up_articulation(art)
				for joint in joint_order:
					joints.append(_dc.find_articulation_dof(art, joint))

			change_collision_at_path(True,paths=['/my_robot_0/camera_link/Cube.physics:collisionEnabled','/my_robot_0/yaw_link/visuals.physics:collisionEnabled'])
			og.Controller.evaluate_sync(_clock_graph)
			# since the first image generated is at time=1/30, we add 7/240
			prev_time = timeline.get_current_time() + 7 / 240 * (simulation_step == 0)
			timeline.set_current_time(prev_time)
			simulation_step += 8

			sleeping(simulation_context, viewport_window_list, is_rtx)
			try:
				if write:
					my_recorder._update()
					my_recorder._enable_record = True
				if write_flow:
					my_recorder_flow._update()
					my_recorder_flow._enable_record = True
				if write_normals:
					my_recorder_normals._update()
					my_recorder_normals._enable_record = True
			except:
				sleeping(simulation_context, viewport_window_list, is_rtx)
				if write:
					my_recorder._update()
					my_recorder._enable_record = True
				if write_flow:
					my_recorder_flow._update()
					my_recorder_flow._enable_record = True
				if write_normals:
					my_recorder_normals._update()
					my_recorder_normals._enable_record = True

			simulation_context.render()
			simulation_context.render()
			timeline.set_current_time(prev_time)
		if simulation_step < 0:
			simulation_context.step(render=False)
			if (simulation_step % ratio_camera == 0):
				timeline.forward_one_frame()
			continue

		if use_teleport:
			if simulation_step % ratio_tf == 0:
				cnt_tf += 1
				teleport("/my_robot_0", np.array(joint_position[cnt_tf][:3]) / meters_per_unit + init_pos
				         , tf.Rotation.from_euler('XYZ', joint_position[cnt_tf][3:] + init_rot).as_quat())
				if (simulation_step % (ratio_tf * 2) == 0):  # odm is published half the rate of the tf
					myp = _dc.get_rigid_body_pose(handle)
					print(
						f"pose diff {np.array(_dc.get_rigid_body_pose(handle).p) / 100 - np.array([robot_pose[int(cnt_tf / 2)][0].x, robot_pose[int(cnt_tf / 2)][0].y, robot_pose[int(cnt_tf / 2)][0].z])}")
			else:
				vel = np.array(joint_velocity[
					               cnt_tf])  # or average position between the two, or use the IMU to interpolate also which has 240 hz
				pos = (np.array(joint_position[cnt_tf][:3]) + vel[:3] * 1 / 240) / meters_per_unit + init_pos
				ori = (np.array(joint_position[cnt_tf][3:]) + vel[3:] * 1 / 240) + init_rot
				teleport("/my_robot_0", pos, tf.Rotation.from_euler('XYZ', ori).as_quat())
		else:
			_dc.wake_up_articulation(art)
			if simulation_step % ratio_tf == 0:
				cnt_tf += 1
				vel = np.array(joint_velocity[cnt_tf])
				next_vel = vel
				if cnt_tf < len(joint_position) - 1:
					next_vel = np.array(joint_velocity[cnt_tf + 1])
				if cnt_tf == 0:
					pos = np.append(np.array(joint_position[cnt_tf][:3]) / meters_per_unit + init_pos - vel[:3] * 1 / 240,
					                joint_position[cnt_tf][3:] + init_rot - vel[3:] * 1 / 240)
					for idx, joint in enumerate(joints):
						_dc.set_dof_position(joint, pos[idx] * (-1 if idx == 1 else 1))

			cvel = (vel + next_vel) / 2
			cvel[:3] = cvel[:3] / meters_per_unit

			_dc.set_articulation_dof_velocity_targets(art, list(cvel))
			for idx, joint in enumerate(joints):
				_dc.set_dof_velocity(joint, cvel[idx] * (-1 if idx == 1 else 1))

			if (simulation_step % (ratio_tf * 2) == 0):
				myp = _dc.get_rigid_body_pose(handle)
				print(
					f"pose diff {np.array(_dc.get_rigid_body_pose(handle).p) / 100 - np.array([robot_pose[int(cnt_tf / 2)][0].x, robot_pose[int(cnt_tf / 2)][0].y, robot_pose[int(cnt_tf / 2)][0].z])}")

				if simulation_step % 8 == 0:
					# tmp = np.load(
					# 	f'/ps/project/irotate/GRADE_DATA/DE/7659a6c9-9fc7-4be5-bc93-5b202ff2a22b/Viewport0/camera/{int(simulation_step/8)}.npy',
					# 	allow_pickle=True).item()
					prim_tf = omni.usd.get_world_transform_matrix(stage.GetPrimAtPath('/my_robot_0/camera_link/Camera'))

		# in v2022 this is the only viable option to control time since timeline.set_auto_update=False is not working
		timeline.set_current_time(prev_time + 1 / 240 * (1 if forward else -1))
		prev_time = timeline.get_current_time()

		simulation_context.step(render=False)
		simulation_context.render()
		print("Clocking...")

		# NOTE THAT THIS MIGHT GET CONFUSING -- reindexing/retiming is needed for sure. Tests need to be careful!
		og.Controller.evaluate_sync(_clock_graph)
		if simulation_step == 0:
			og.Controller.evaluate_sync(_clock_graph)
		time.sleep(0.2)


		if simulation_step % ratio_camera == 0:
			if (simulation_step + ratio_camera) / ratio_camera < (experiment_length / reversing_timeline_ratio) * (
					cnt_reversal):
				forward = True
			else:
				if (simulation_step + ratio_camera) / ratio_camera >= ((experiment_length - 1) / reversing_timeline_ratio) * (
						cnt_reversal + 1) or \
						(timeline.get_current_time() - 1 / timeline.get_time_codes_per_seconds()) < 0:
					cnt_reversal += 2
					forward = True
				else:
					forward = False
			if write_flow:
				if my_recorder_flow._enable_record:
					simulation_context.render()
					my_recorder_flow._counter += 1
					time.sleep(1.5)  # this seems necessary
					my_recorder_flow._update()

			# you have two ways to proceed here. the sleeping performs just the rendering and then you manually toggle the recorder below
			# otherwise use pub_and_write_images which automatically calls it if necessary. In the latter case, remember to increase the counter
			sleeping(simulation_context, viewport_window_list, is_rtx)
			# if write:
			# 	if my_recorder._enable_record:
			# 		my_recorder._counter += 1
			# pub_and_write_images(simulation_context, viewport_window_list, ros_camera_list, is_rtx, my_recorder)

			if write:
				if my_recorder._enable_record:
					my_recorder._counter += 1
					my_recorder._update()
			if write_normals:
				if my_recorder_normals._enable_record:
					my_recorder_normals._counter += 1
					my_recorder_normals._update()

		# new sensor here -- imagine 30 fps -- in that case I need to publish
		# if you need sensors in the middle you need to interpolate
		# using IMU and TF readings
		# you can access those from the rosbags
		# note you might need to work with the timeline times if the rate that you want is different
		# if simulation_step % ratio_camera == 0:
		# 	for lidar in lidars:
		# 		og.Controller.attribute(lidar + ".inputs:step").set(1)
		# 	ctime = timeline.get_current_time()
		# 	simulation_context.render()
		# 	# point_cloud = og.Controller().node("/Render/PostProcess/SDGPipeline/RenderProduct_Replicator_RtxSensorCpuIsaacComputeRTXLidarPointCloud").get_attribute("outputs:pointCloudData").get()
		# 	# laser_scan = og.Controller().node("/Render/PostProcess/SDGPipeline/RenderProduct_Replicator_RtxSensorCpuIsaacComputeRTXLidarFlatScan").get_attribute("outputs:linearDepthData").get()
		# 	timeline.set_current_time(ctime)
		# 	for lidar in lidars:
		# 		og.Controller.attribute(lidar+".inputs:step").set(0)

		if simulation_step % ratio_camera == 0 and simulation_step / ratio_camera == experiment_length:
			print("End of experiment!!!")
			simulation_context.pause()
			break
except:
	extype, value, tb = sys.exc_info()
	traceback.print_exc()

	import ipdb

	ipdb.set_trace()
finally:
	simulation_context.stop()
try:
	kit.close()
except:
	pass