Skip to content

srli/jimmy

 
 

Repository files navigation

1. Overview:
We decompose the motion control problem into two parts, a high level 
controller that generates trajectories for specific body parts in the Cartesian
space, e.g. move foot or hand around, and a low level controller that solves 
Inverse Kinematics (IK) to compute all the joint trajectories that best fit the 
high level controller's goal. Since the low level controller is generic, this 
approach facilitates fast development for multiple behaviors. We used the same
low level controller for both walking and gesturing, as they only differ in
the high level controller. Due to limited communication bandwidth with the 
servos, we are do not use any feedback. Perfect servo tracking is assumed. 

1.1 Inverse Kinematics:
Different from most IK approaches that solve joint positions for the entire 
trajectory a priori, we solve for joint velocity on each tick and then 
integrate to get joint positions. This way, we can will get smooth joint 
trajectories. It also responds fast, and allows the high level controller to 
change objectives midway through a motion. We formulate the IK problem as a
Quadratic Programming problem, which minimizes a quadratic cost subject to 
linear constraints. The unknown for QP consists of all the joint velocities 
(20, without the neck joints) and six velocities for the floating pelvis (qdot). 
The quadratic cost mostly penalizes center of mass, hand, foot or individual 
joints velocities from the desired values given by the high level controller. 
Joint limits are treated as joint limits. On every tick, we solve this QP to 
get a new qdot, and use it to integrate to get a new position (q) for all the 
joint position and pelvis position and orientation. The joint positions are 
passed to the servo as targets.

1.2 Walking:
The operator will specify forward and turning velocities for walking, and the 
high level controller will generate a sequence of foot steps which are then 
used to generate swing foot and Center of Mass (COM) trajectories. These 
trajectories are computed using a Linear Inverted Pendulum (LIPM) model, which 
is a common approach for flat floor static walking. These trajectories are then
passed to inverse kinematics. 

1.3 Gesturing:
In gesturing mode, the high level controller generates trajectories by a 
quintic spline with the current states and stored desired states as knot
points. The user have control over all 8 arm joints, torso orientation (RPY), 
pelvis position relative the to the middle of the foot (XYZ) and 3 neck joints. 
IK will compute the leg joints, using pelvis position and torso orientation 
instead of leg joints makes scripting / puppetry easier. 

1.4 Scripting / puppetry
You can run bin/test_puppet to start puppetry. The arms and neck are loose.
After you pose the robot, hit Enter to save all the current joint positions in
a file in conf/puppet. The positions are appended to the end of the file. The
order of joints: 6 left leg, 6 right leg, 4 left arm, 4 right arm, 3 neck. 
For specific joint, check include/jimmy/JimmyCommon.h under Joint. For the 3 
letter encoding, the last 2 stands for the type of joint. FE is pitch, Z is yaw, 
AA is roll, where X axis points forward, Y points to the left, and Z points up.
The first letter encodes which joint, H for hip, K for knee, A for ankle, S for
shoulder, E for elbow. 

1.5 Servo communication
We use USB2Dynamixel usb adaptor to talk to the servos. The servo supports sync 
write instruction but not sync read, which means we can write to a large number 
of servos in 1 big packet (fast), but have to sequentially query individual
servo for feedback (slow). The current implementation writes target position to 
all servos at 100 Hz, but only reads encoders from 2 servos in a round robin 
way. 

2. config files format. 
conf/
  IK.cf: weights for IK
  plan.cf: parameters for walking, i.e. swing length, step size etc
  poses.cf: pose file. Each line specifies a desired pose. 
            First 8 numbers are arm joint angles. 
            Next 3 are torso RPY.
            Next 3 are pelvis position (XYZ) relative to the middle of 
              the feet.
            Next 3 head angles. 
  gestures.cf: Each line represents a gesture, which is a list of two 
            numbers. This first number is the index to the line number in 
            poses.cf (0 index), and the second number is the length in 
            seconds. 
  puppet: temp file that stores the logged joint positions when in puppetry 
          mode. 

3 ROS messages
The message definition is in msg/jimmy_command.msg
The message has 1 int for its type, and variable number of doubles parameters 
that are used for neck motion or steering walking. 


4. Examples
4.1 bin/test_puppet 
For puppetry. Press Enter to save a pose.

4.2 bin/test_jimmy 
Main code. Starts the robot in standing mode. The user can send ROS messages
to change mode or give commands. 

About

No description, website, or topics provided.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages

  • C 60.9%
  • C++ 30.3%
  • Makefile 4.7%
  • Java 1.1%
  • Python 1.0%
  • Processing 0.9%
  • Other 1.1%