Warning
This tutorial requires a recent version of MORSE. Follow the instructions here to install the latest version.
Note
This tutorial will guide you through the development of a complete simulation, including the creation of a robot and a new actuators from scratch.
To start this journey, you need a working installation of MORSE. If it is not the case, refer to the installation notes first, and come back after!
Let’s start by giving a destination to our journey: the Ranger robot.
The Ranger is a robot developped at EPFL, at the CHILI and LSRO labs. The project aims at creating a nice, funny robot that would help children to tidy up their rooms.
It is basically a wooden box, with two wheels, animated eyes. It can play sounds and have a large amount of LEDs on the faces.
After spending some minutes in the MORSE component library, it appears that the Differential Driver Actuator: Linear and angular speed (V, W) actuator may be a good candidate to control the wheels.
However, for the eyes and the LEDs, we will need to create something new.
But first, we want to create a 3D model for our robot.
Making a new robot model does not require anything beyond creating a 3D mesh in Blender (which also mean you can import existing meshes in a variety of format in Blender).
Note
Creating a realistic robot model may become more complex, especially if you want/need to use armatures to represent complex kinematic chains, or if you want to create low-poly bounding boxes for good performances when simulating collisions.
In the MORSE data directory (typically, /usr/local/share/morse/data), many 3D models are available that can be good starting points.
In our case, props/crates.blend contains 3 models of wooden box that we can use as starting point.
But first, let’s create a new MORSE project.
Choose a location where you would like to store your simulation, and run:
$ morse create ranger_sim
This creates a new directory called ranger_sim, with a file called default.py, and two sub-directories: scripts/ and src/ (currently empty). scripts/ contains a sample test client, we come back to that in chapter 3.
We can straight away test our simulation:
$ morse run ranger_sim
After MORSE has loaded, it should display something similar to this:
You can already move around the robot (Morsy, the MORSE mascot!) with the arrow keys, and it should create collision if you drive it to an obstacle.
Note
You can control the camera by maintaining down the Ctrl key and moving around the mouse, and with the WASD keys.
Let’s now create our own robot model.
First, ask MORSE to create for us the templates for a new robot called ranger inside our ranger_sim simulation:
$ morse add robot ranger ranger_sim
This create several new files in ranger_sim/: data/ranger_sim/robots/ranger.blend contains a default mesh for our robot, src/ranger_sim/robots/ranger.py describes the behaviour of our simulated robot (currently, it does nothing particular) and src/ranger_sim/builder/robots/ranger.py describes the equipment of the Ranger. Initially, a simple motion controller and a position sensor.
To use this new robot in our simulation, open ranger_sim/default.py with your favorite editor, and replace the default robot with the Ranger: add from ranger_sim.builder.robots import Ranger at the top of your file, and replace on line 15 Morsy by Ranger. You can also remove the lines that add the motion controller, the keyboard controller and the pose sensor since these are already part of the default equipment of our robot.
The new default.py should look like that:
from morse.builder import *
from ranger_sim.builder.robots import Ranger
robot = Ranger()
robot.translate(1.0, 0.0, 0.0)
robot.add_default_interface('socket')
env = Environment('sandbox', fastmode = False)
env.set_camera_location([10.0, -10.0, 10.0])
env.set_camera_rotation([1.05, 0, 0.78])
Run again this simulation with:
$ morse run ranger_sim
You should see... no differences! The robot template created by MORSE uses the same mesh as the Morsy robot. We will fix that now.
Open the current default mesh of the Ranger with Blender:
$ cd ranger_sim
$ blender data/ranger_sim/robots/ranger.blend
Blender opens and should display the Morsy wireframe:
You can press z with the mouse over the 3D view to switch from the wireframe to the solid model.
Now:
Here, the object has been imported on the third layer. Click it, select everything (a), then move it to first layer (m), and switch back to the first layer.
It looks better, but we can improve it even more.
Re-open ranger.blend in Blender and follow these steps:
Warning
In MORSE, the X axis is the forward axis: that is not the Blender convention. You want to add eyes on the Right Ortho face in Blender interface.
Save you file, quit Blender and run your simulation: that should be enough to start playing a bit around!
In the next chapter, we’ll see how to control the movement of the robot from an external application (your robot controller), and in the following chapter, we will try to move the eyes (we will have to create a new dedicated actuator to this end).
MORSE is all about integrating simulation as transparently as possible in your usual workflow: we want you to be able to switch almost transparently from a real robot to the simulated robot.
To achieve that, MORSE relies on intermediate middlewares, like ROS, YARP, etc. You can get an idea of which features of each middlewares are supported by MORSE on this page.
Sometimes, for quick prototyping, or simply to evaluate what is actually possible to simulate, you do not want to use a full middleware, but a more lightweight interface. MORSE comes with a simple socket-based interface for that purpose, and also a simple-to-use Python binding that will enable us to quickly test our robot controller.
Let write a first test right away. Open your favorite editor and copy-paste this example:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 | from pymorse import Morse
print("Use WASD to control the Ranger")
with Morse() as simu:
motion = simu.robot.motion
v = 0.0
w = 0.0
while True:
key = input("WASD?")
if key.lower() == "w":
v += 0.1
elif key.lower() == "s":
v -= 0.1
elif key.lower() == "a":
w += 0.1
elif key.lower() == "d":
w -= 0.1
else:
continue
motion.publish({"v": v, "w": w})
|
Save it somewhere (for instance, as ranger_sim/scripts/keyboard_ctrl.py), launch the simulation ($ morse run ranger_sim) and run your script:
$ morse run ranger_sim &
$ python3 ranger_sim/scripts/keyboard_ctrl.py
Note
A client script, very similar to this one, is automatically generated when creating the simulation environment. You will find it under scripts/ranger_sim_client.py.
Warning
pymorse the Python bindings for MORSE, require Python 3
Press Ctrl + c to interrupt the keyboard_ctrl.py script, and let examine the detail of this first example.
On line 1, we import the pymorse bindings. To use them, we create on line 5 a context: at creation, the connection is established with the simulator (by default, on localhost, but you can change that), and when we exit the context, the connections are properly closed. In this example, the context object is stored in the simu variable.
On line 7, we retrieve the motion controller end-point. The names to access it is the same as the way we named our components in the simulation script (default.py and ranger.py).
Note
Here, in default.py, on line 5, we called our robot robot by simply naming that way the Ranger() object:
1 2 3 4 5 6 7 8 9 10 11 | from morse.builder import *
from robots import Ranger
robot = Ranger()
# The list of the main methods to manipulate your components
# is here: http://www.openrobots.org/morse/doc/stable/user/builder_overview.html
robot.translate(1.0, 0.0, 0.0)
[...]
|
If you check src/ranger_sim/builder/robots/ranger.py, you will find out that the Ranger’s motion controller has been called motion. So with pymorse, we access the motion controller simply as simu.robot.motion.
Then, at line 12, we start the main loop: we read a keyboard input, we change the linear v and radial w speeds depending on the user input, and, line 26, we send to the simulator the new command.
The command is a plain Python dictionary, whose content depends on the actuator. In our case, we are using a MotionVW actuator (see src/ranger_sim/builder/robots/ranger.py). The documentation of the component tells us what the actuator expects.
motion is an actuator. If you open robots/ranger.py, you will see the template also declare a Pose sensor. We can access it to print the current position of the robot. Open scripts/keyboard_ctrl.py and modify it that way:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 | from pymorse import Morse
def pose_received(pose):
print("The Ranger is currently at %s" % pose)
print("Use WASD to control the Ranger")
with Morse() as simu:
simu.robot.pose.subscribe(pose_received)
motion = simu.robot.motion
v = 0.0
w = 0.0
while True:
key = input("WASD?")
if key.lower() == "w":
v += 0.1
elif key.lower() == "s":
v -= 0.1
elif key.lower() == "a":
w += 0.1
elif key.lower() == "d":
w -= 0.1
else:
continue
motion.publish({"v": v, "w": w})
|
Restart the scripts/keyboard_ctrl.py script:
$ python3 ranger_sim/scripts/keyboard_ctrl.py
It should start quickly filling your console with the position of the robot. You can still control it with WASD as you did previously, and you should see the position values changing.
Let now create a new custom actuator for the eyes of our robot.
Add a new actuator template called eyes to the ranger_sim simulation:
$ morse add actuator eyes ranger_sim
MORSE asks you for a short description of your actuator (enter something like “Controls the eyes of the EPFL Ranger robot”) , and then create a new set of templates: src/actuators/eyes.py defines the behaviour of the actuator (how the actuator interacts with the simulation) and src/builder/actuators/eyes.py provides the Builder API interface to use the actuator in simulation scripts.
Note
Contrary to robots, where you are encouraged to modify their Builder API to define the robot equipment, you usually do not need to change it for actuators (or sensors).
The only case where it may be useful is to specify a special 3D mesh for your component (like the casing of a laser scanner, etc.)
The default actuator template does not provide any useful behaviour, but it can already be added to our robot:
from ranger_sim.builder.actuators import Eyes
and these two lines after the motion controller:
self.eyes = Eyes()
self.append(self.eyes)
If you launch the simulation now, MORSE will list the components available on our robot, including the eyes:
[...]
[ 0.283] ------------------------------------
[ 0.284] - SIMULATION SUMMARY -
[ 0.284] ------------------------------------
[ 0.284] Robots in the simulation:
[ 0.284] ROBOT: 'robot'
[ 0.284] - Component: 'robot.pose'
[ 0.285] - Component: 'robot.keyboard'
[ 0.285] - Component: 'robot.eyes'
[ 0.285] - Component: 'robot.motion'
[...]
For our actuator to produce something tangible, we need to complete its definition.
Open src/ranger_sim/actuators/eyes.py, and update its content to match the following Python script:
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 | import logging; logger = logging.getLogger("morse." + __name__)
from morse.core.actuator import Actuator
from morse.helpers.components import add_data
from morse.core import mathutils
class Eyes(Actuator):
_name = "Eyes"
_short_desc = "Controls the eyes of the EPFL Ranger robot"
add_data('left', 0.1, 'float', 'Left eye rotation, in radians')
add_data('right', -0.1, 'float', 'Right eye rotation, in radians')
def __init__(self, obj, parent=None):
logger.info("%s initialization" % obj.name)
# Call the constructor of the parent class
Actuator.__init__(self, obj, parent)
self.left_eye = parent.bge_object.children["left_eye"]
self.right_eye = parent.bge_object.children["right_eye"]
logger.info('Component initialized')
def default_action(self):
l_orientation = mathutils.Euler([self.local_data['left'], 0.0, 0.0])
self.left_eye.orientation = l_orientation.to_matrix()
r_orientation = mathutils.Euler([self.local_data['right'], 0.0, 0.0])
self.right_eye.orientation = r_orientation.to_matrix()
|
Let’s explain this script:
class Eyes(Actuator):
_name = "Eyes"
_short_desc = "Controls the eyes of the EPFL Ranger robot"
When creating a component, always provide a name and short description. These are used to generate the component documentation for instance.
add_data('left', 0.1, 'float', 'Left eye rotation, in radians')
add_data('right', -0.1, 'float', 'Right eye rotation, in radians')
These two lines define the data interface of our actuator. For the eyes, we need to provide to the actuator two angles, one per eye.
We first set the name of the data field, then its default value, its type and a short description.
The data set by the simulator clients can be later accessed through the local_data dictionary (see below).
def __init__(self, obj, parent=None):
logger.info("%s initialization" % obj.name)
# Call the constructor of the parent class
Actuator.__init__(self, obj, parent)
self.left_eye = parent.bge_object.children["left_eye"]
self.right_eye = parent.bge_object.children["right_eye"]
logger.info('Component initialized')
The class constructor has nothing special. self.left_eye and self.right_eye are set to point to the Blender objects for the eyes (parent is the robot body, parent.bge_object represents the Blender mesh of the robot body, parent.bge_object.children contains all children of the robot mesh).
def default_action(self):
l_orientation = mathutils.Euler([self.local_data['left'], 0.0, 0.0])
self.left_eye.orientation = l_orientation.to_matrix()
r_orientation = mathutils.Euler([self.local_data['right'], 0.0, 0.0])
self.right_eye.orientation = r_orientation.to_matrix()
default_action() is the most important method of a component. It is called at each simulation step. The behaviour of the actuator is implemented here.
For our eyes, we simply apply a rotation along the X axis (Blender uses rotation matrices., so we first create the rotation matrix from a vector of Euler angles).
To test the eyes, we must complete our test client.
Re-open scripts/keyboard_ctrl.py, and update it this way:
from pymorse import Morse
print("Use WASD to control the Ranger")
with Morse() as simu:
motion = simu.robot.motion
eyes = simu.robot.eyes
v = 0.0
w = 0.0
left = 0.0
right = 0.0
while True:
key = input("WASD (eyes:RFTG)?")
if key.lower() == "w":
v += 0.1
elif key.lower() == "s":
v -= 0.1
elif key.lower() == "a":
w += 0.1
elif key.lower() == "d":
w -= 0.1
elif key.lower() == "r":
left += 0.1
elif key.lower() == "f":
left -= 0.1
elif key.lower() == "t":
right += 0.1
elif key.lower() == "g":
right -= 0.1
else:
continue
motion.publish({"v": v, "w": w})
eyes.publish({"left": left, "right": right})
Besides (v, ω), we now also publish on the eyes channel a pair (left, right).
Run the simulation and launch your client:
$ morse run ranger_sim &
$ python3 ranger_sim/scripts/keyboard_ctrl.py
You should now be able to move the eyes:
To be done!
To be done!