PyBullet is an open-source simulation platform created by Fb that’s designed for coaching bodily brokers (corresponding to robots) in a 3D atmosphere. It supplies a physics engine for each inflexible and gentle our bodies. It’s generally used for robotics, synthetic intelligence, and sport improvement.
Robotic arms are highly regarded on account of their pace, precision, and talent to work in hazardous environments. They’re used for duties corresponding to welding, meeting, and materials dealing with, particularly in industrial settings (like manufacturing).
There are two methods for a robotic to carry out a process:
- Guide Management – requires a human to explicitly program each motion. Higher fitted to fastened duties, however struggles with uncertainty and requires tedious parameter tuning for each new situation.
- Synthetic Intelligence – permits a robotic to be taught the perfect actions by way of trial and error to attain a purpose. So, it may adapt to altering environments by studying from rewards and penalties with out a predefined plan (Reinforcement Studying).
On this article, I’m going to point out the way to construct a 3D atmosphere with PyBullet for manually controlling a robotic arm. I’ll current some helpful Python code that may be simply utilized in different comparable instances (simply copy, paste, run) and stroll by way of each line of code with feedback so that you could replicate this instance.
Setup
PyBullet can simply be put in with one of many following instructions (if pip fails, conda ought to undoubtedly work):
pip set up pybullet
conda set up -c conda-forge pybulletPyBullet comes with a set of preset URDF information (Unified Robotic Description Format), that are XML schemas describing the bodily construction of objects within the 3D atmosphere (i.e. dice, sphere, robotic).
import pybullet as p
import pybullet_data
import time
## setup
p.join(p.GUI)
p.resetSimulation()
p.setGravity(gravX=0, gravY=0, gravZ=-9.8)
p.setAdditionalSearchPath(path=pybullet_data.getDataPath())
## load URDF
airplane = p.loadURDF("airplane.urdf")
dice = p.loadURDF("dice.urdf", basePosition=[0,0,0.1], globalScaling=0.5, useFixedBase=True)
obj = p.loadURDF("cube_small.urdf", basePosition=[1,1,0.1], globalScaling=1.5)
p.changeVisualShape(objectUniqueId=obj, linkIndex=-1, rgbaColor=[1,0,0,1]) #purple
whereas p.isConnected():
p.setRealTimeSimulation(True)
Let’s undergo the code above:
- when you’ll be able to connect with the physics engine, it’s worthwhile to specify if you wish to open the graphic interface (
p.GUI) or not (p.DIRECT). - The Cartesian house has 3 dimensions: x-axis (ahead/backward), y-axis (left/proper), z-axis (up/down).
- It’s widespread apply to set the gravity to
(x=0, y=0, z=-9.8)simulating Earth’s gravity, however one can change it based mostly on the target of the simulation. - Often, it’s worthwhile to import a airplane to create a floor, in any other case objects would simply fall indefinitely. In order for you an object to be fastened to the ground, then set
useFixedBase=True(it’s False by default). I imported the objects withbasePosition=[0,0,0.1]that means that they’re 10cm above the bottom. - The simulation will be rendered in real-time with
p.setRealTimeSimulation(True)or sooner (CPU-time) withp.stepSimulation(). One other approach to set real-time is:
import time
for _ in vary(240): #240 timestep generally utilized in videogame improvement
p.stepSimulation() #add a physics step (CPU pace = 0.1 second)
time.sleep(1/240) #decelerate to real-time (240 steps × 1/240 second sleep = 1 second)Robotic
At the moment, our 3D atmosphere is made from a little bit object (tiny purple dice), and a desk (large dice) fastened to the bottom (airplane). I shall add a robotic arm with the duty of selecting up the smaller object and placing it on the desk.
PyBullet has a default robotic arm modeled after the Franka Panda Robotic.
robo = p.loadURDF("franka_panda/panda.urdf",
basePosition=[1,0,0.1], useFixedBase=True)
The very first thing to do is to investigate the hyperlinks (the strong components) and joints (connections between two inflexible components) of the robotic. For this function, you’ll be able to simply use p.DIRECT as there isn’t a want for the graphic interface.
import pybullet as p
import pybullet_data
## setup
p.join(p.DIRECT)
p.resetSimulation()
p.setGravity(gravX=0, gravY=0, gravZ=-9.8)
p.setAdditionalSearchPath(path=pybullet_data.getDataPath())
## load URDF
robo = p.loadURDF("franka_panda/panda.urdf",
basePosition=[1,0,0.1], useFixedBase=True)
## joints
dic_info = {
0:"joint Index", #begins at 0
1:"joint Identify",
2:"joint Kind", #0=revolute (rotational), 1=prismatic (sliding), 4=fastened
3:"state vectorIndex",
4:"velocity vectorIndex",
5:"flags", #nvm all the time 0
6:"joint Damping",
7:"joint Friction", #coefficient
8:"joint lowerLimit", #min angle
9:"joint upperLimit", #max angle
10:"joint maxForce", #max drive allowed
11:"joint maxVelocity", #max pace
12:"hyperlink Identify", #little one hyperlink related to this joint
13:"joint Axis",
14:"guardian FramePos", #place
15:"guardian FrameOrn", #orientation
16:"guardian Index" #−1 = base
}
for i in vary(p.getNumJoints(bodyUniqueId=robo)):
joint_info = p.getJointInfo(bodyUniqueId=robo, jointIndex=i)
print(f"--- JOINT {i} ---")
print({dic_info[k]:joint_info[k] for ok in dic_info.keys()})
## hyperlinks
for i in vary(p.getNumJoints(robo)):
link_name = p.getJointInfo(robo, i)[12].decode('utf-8') #area 12="hyperlink Identify"
dyn = p.getDynamicsInfo(robo, i)
pos, orn, *_ = p.getLinkState(robo, i)
dic_info = {"Mass":dyn[0], "Friction":dyn[1], "Place":pos, "Orientation":orn}
print(f"--- LINK {i}: {link_name} ---")
print(dic_info)
Each robotic has a “base“, the basis a part of its physique that connects all the things (like our skeleton backbone). Wanting on the output of the code above, we all know that the robotic arm has 12 joints and 12 hyperlinks. They’re related like this:
Motion Management
With the intention to make a robotic do one thing, you need to transfer its joints. There are 3 primary forms of management, that are all purposes of Newton’s legal guidelines of movement:
- Place management: principally, you inform the robotic “go right here”. Technically, you set a goal place, after which apply drive to regularly transfer the joint from its present place towards the goal. Because it will get nearer, the utilized drive decreases to keep away from overshooting and finally balances completely in opposition to resistive forces (i.e. friction, gravity) to carry the joint regular in place.
- Velocity management: principally, you inform the robotic “transfer at this pace”. Technically, you set a goal velocity, and apply drive to regularly deliver the speed from zero to the goal, then keep it by balancing utilized drive and resistive forces (i.e. friction, gravity).
- Pressure/Torque management: principally, you “push the robotic and see what occurs”. Technically, you instantly set the drive utilized on the joint, and the ensuing movement relies upon purely on the article’s mass, inertia, and the atmosphere. As a aspect word, in physics, the phrase “drive” is used for linear movement (push/pull), whereas “torque” signifies rotational movement (twist/flip).
We are able to use p.setJointMotorControl2() to maneuver a single joint, and p.setJointMotorControlArray() to use drive to a number of joints on the identical time. As an illustration, I shall carry out place management by giving a random goal for every arm joint.
## setup
p.join(p.GUI)
p.resetSimulation()
p.setGravity(gravX=0, gravY=0, gravZ=-9.8)
p.setAdditionalSearchPath(path=pybullet_data.getDataPath())
## load URDF
airplane = p.loadURDF("airplane.urdf")
dice = p.loadURDF("dice.urdf", basePosition=[0,0,0.1], globalScaling=0.5, useFixedBase=True)
robo = p.loadURDF("franka_panda/panda.urdf", basePosition=[1,0,0.1], useFixedBase=True)
obj = p.loadURDF("cube_small.urdf", basePosition=[1,1,0.1], globalScaling=1.5)
p.changeVisualShape(objectUniqueId=obj, linkIndex=-1, rgbaColor=[1,0,0,1]) #purple
## transfer arm
joints = [0,1,2,3,4,5,6]
target_positions = [1,1,1,1,1,1,1] #<--random numbers
p.setJointMotorControlArray(bodyUniqueId=robo, jointIndices=joints,
controlMode=p.POSITION_CONTROL,
targetPositions=target_positions,
forces=[50]*len(joints))
for _ in vary(240*3):
p.stepSimulation()
time.sleep(1/240)
The true query is, “what’s the proper goal place for every joint?” The reply is Inverse Kinematics, the mathematical strategy of calculating the parameters wanted to position a robotic in a given place and orientation relative to a place to begin. Every joint defines an angle, and also you don’t need to guess a number of joint angles by hand. So, we’ll ask PyBullet to determine the goal positions within the Cartesian house with p.calculateInverseKinematics().
obj_position, _ = p.getBasePositionAndOrientation(obj)
obj_position = listing(obj_position)
target_positions = p.calculateInverseKinematics(
bodyUniqueId=robo,
endEffectorLinkIndex=11, #grasp-target hyperlink
targetPosition=[obj_position[0], obj_position[1], obj_position[2]+0.25], #25cm above object
targetOrientation=p.getQuaternionFromEuler([0,-3.14,0]) #[roll,pitch,yaw]=[0,-π,0] -> hand pointing down
)
arm_joints = [0,1,2,3,4,5,6]
p.setJointMotorControlArray(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndices=arm_joints,
targetPositions=target_positions[:len(arm_joints)],
forces=[50]*len(arm_joints))
Please word that I used p.getQuaternionFromEuler() to convert the 3D angles (simple for people to grasp) into 4D, as “quaternions” are simpler for physics engines to compute. If you wish to get technical, in a quaternion (x, y, z, w), the primary three elements describe the axis of rotation, whereas the fourth dimension encodes the quantity of rotation (cos/sin).
The code above instructions the robotic to maneuver its hand to a particular place above an object utilizing Inverse Kinematics. We seize the place the little purple dice is sitting on this planet with p.getBasePositionAndOrientation(), and we use the knowledge to calculate the goal place for the joints.
Work together with Objects
The robotic arm has a hand (“gripper”), so it may be opened and closed to seize objects. From the earlier evaluation, we all know that the “fingers” can transfer inside (0-0.04). So, you’ll be able to set the goal place because the decrease restrict (hand closed) or the higher restrict (hand open).
Furthermore, I need to be sure that the arm holds the little purple dice whereas shifting round. You should use p.createConstraint() to make a momentary physics bond between the robotic’s gripper and the article, so that it’s going to transfer along with the robotic hand. In the actual world, the gripper would apply drive by way of friction and get in touch with to squeeze the article so it doesn’t fall. However, since PyBullet is a quite simple simulator, I’ll simply take this shortcut.
## shut hand
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=9, #finger_joint1
targetPosition=0, #decrease restrict for finger_joint1
drive=drive)
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=10, #finger_joint2
targetPosition=0, #decrease restrict for finger_joint2
drive=drive)
## maintain the article
constraint = p.createConstraint(
parentBodyUniqueId=robo,
parentLinkIndex=11,
childBodyUniqueId=obj,
childLinkIndex=-1,
jointType=p.JOINT_FIXED,
jointAxis=[0,0,0],
parentFramePosition=[0,0,0],
childFramePosition=[0,0,0,1]
)After that, we are able to transfer the arm towards the desk, utilizing the identical technique as earlier than: Inverse Kinematics -> goal place -> place management.
Lastly, when the robotic reaches the goal place within the Cartesian house, we are able to open the hand and launch the constraint between the article and the arm.
## shut hand
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=9, #finger_joint1
targetPosition=0.04, #higher restrict for finger_joint1
drive=drive)
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=10, #finger_joint2
targetPosition=0.04, #higher restrict for finger_joint2
drive=drive)
## drop the obj
p.removeConstraint(constraint)
Full Guide Management
In PyBullet, it’s worthwhile to render the simulation for each motion the robotic takes. Subsequently, it’s handy to have a utils perform that may pace up (i.e. sec=0.1) or decelerate (i.e. sec=2) the real-time (sec=1).
import pybullet as p
import time
def render(sec=1):
for _ in vary(int(240*sec)):
p.stepSimulation()
time.sleep(1/240)Right here’s the complete code for the simulation we now have performed on this article.
import pybullet_data
## setup
p.join(p.GUI)
p.resetSimulation()
p.setGravity(gravX=0, gravY=0, gravZ=-9.8)
p.setAdditionalSearchPath(path=pybullet_data.getDataPath())
airplane = p.loadURDF("airplane.urdf")
dice = p.loadURDF("dice.urdf", basePosition=[0,0,0.1], globalScaling=0.5, useFixedBase=True)
robo = p.loadURDF("franka_panda/panda.urdf", basePosition=[1,0,0.1], globalScaling=1.3, useFixedBase=True)
obj = p.loadURDF("cube_small.urdf", basePosition=[1,1,0.1], globalScaling=1.5)
p.changeVisualShape(objectUniqueId=obj, linkIndex=-1, rgbaColor=[1,0,0,1])
render(0.1)
drive = 100
## open hand
print("### open hand ###")
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=9, #finger_joint1
targetPosition=0.04, #higher restrict for finger_joint1
drive=drive)
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=10, #finger_joint2
targetPosition=0.04, #higher restrict for finger_joint2
drive=drive)
render(0.1)
print(" ")
## transfer arm
print("### transfer arm ### ")
obj_position, _ = p.getBasePositionAndOrientation(obj)
obj_position = listing(obj_position)
target_positions = p.calculateInverseKinematics(
bodyUniqueId=robo,
endEffectorLinkIndex=11, #grasp-target hyperlink
targetPosition=[obj_position[0], obj_position[1], obj_position[2]+0.25], #25cm above object
targetOrientation=p.getQuaternionFromEuler([0,-3.14,0]) #[roll,pitch,yaw]=[0,-π,0] -> hand pointing down
)
print("goal place:", target_positions)
arm_joints = [0,1,2,3,4,5,6]
p.setJointMotorControlArray(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndices=arm_joints,
targetPositions=target_positions[:len(arm_joints)],
forces=[force/3]*len(arm_joints))
render(0.5)
print(" ")
## shut hand
print("### shut hand ###")
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=9, #finger_joint1
targetPosition=0, #decrease restrict for finger_joint1
drive=drive)
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=10, #finger_joint2
targetPosition=0, #decrease restrict for finger_joint2
drive=drive)
render(0.1)
print(" ")
## maintain the article
print("### maintain the article ###")
constraint = p.createConstraint(
parentBodyUniqueId=robo,
parentLinkIndex=11,
childBodyUniqueId=obj,
childLinkIndex=-1,
jointType=p.JOINT_FIXED,
jointAxis=[0,0,0],
parentFramePosition=[0,0,0],
childFramePosition=[0,0,0,1]
)
render(0.1)
print(" ")
## transfer in direction of the desk
print("### transfer in direction of the desk ###")
cube_position, _ = p.getBasePositionAndOrientation(dice)
cube_position = listing(cube_position)
target_positions = p.calculateInverseKinematics(
bodyUniqueId=robo,
endEffectorLinkIndex=11, #grasp-target hyperlink
targetPosition=[cube_position[0], cube_position[1], cube_position[2]+0.80], #80cm above the desk
targetOrientation=p.getQuaternionFromEuler([0,-3.14,0]) #[roll,pitch,yaw]=[0,-π,0] -> hand pointing down
)
print("goal place:", target_positions)
arm_joints = [0,1,2,3,4,5,6]
p.setJointMotorControlArray(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndices=arm_joints,
targetPositions=target_positions[:len(arm_joints)],
forces=[force*3]*len(arm_joints))
render()
print(" ")
## open hand and drop the obj
print("### open hand and drop the obj ###")
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=9, #finger_joint1
targetPosition=0.04, #higher restrict for finger_joint1
drive=drive)
p.setJointMotorControl2(bodyUniqueId=robo, controlMode=p.POSITION_CONTROL,
jointIndex=10, #finger_joint2
targetPosition=0.04, #higher restrict for finger_joint2
drive=drive)
p.removeConstraint(constraint)
render()
Conclusion
This text has been a tutorial on the way to manually management a robotic arm. We discovered about motion management, Inverse Kinematics, grabbing and shifting objects. New tutorials with extra superior robots will come.
Full code for this text: GitHub
I hope you loved it! Be at liberty to contact me for questions and suggestions, or simply to share your attention-grabbing initiatives.
👉 Let’s Join 👈
(All photos are by the creator until in any other case famous)
