Differential Drive Mobile Robot

From Mech
Revision as of 08:10, 27 April 2010 by JarvisSchultz (talk | contribs) (→‎Code Download)
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)
Jump to navigationJump to search


Mobile Robot - Parametric - Small.jpg


The goal of this project was to create a low cost replacement for the popular E-Puck robot that was better suited to the needs of current research projects here at Northwestern. The project also entailed developing a control library that would allow future users to easily interface with the robot and integrate it into their research. Specifically, the robot was to meet the following criteria:

Mechanical Goals

  • Be of a similar size to the E-Puck
  • Be significantly less expensive than the E-Puck
  • Use stepper motors
  • Be easy to manufacture
  • Have an approximate run time of 1 hour

Software Goals

  • Use the NU32 board for control logic
  • Support wireless communication
  • Develop a control library that can easily be included in a users code
  • Allow the user to specify a velocity vector for the robot to follow
  • Allow the user to specify the robot's configuration, (x,y,θ), in a global frame
  • Allow the user to directly control the robot's wheel speeds

Demonstration Video

A video of the robot's operation can be viewed here


Hybrid Stepper Motor

As previously stated, the robot needed to be easy to manufacture, inexpensive to make, and relatively small. These three goals guided each decision made in the mechanical design. The following sections will outline the design process for each mechanical component.


The motor choice was the most significant design decision in this project. Accordingly, much thought and research went into picking the hybrid stepper motors we ended up with. The stepper motor used in this project is a NEMA 17, bipolar (4 wire), hybrid stepper. An image of this can be seen on the right.

  • Standard Size:
Very early in the motor search, it was decided that because this project was to be the foundation of future work, finding a motor that was readily available was a top priority. It quickly became clear that using a stepper motor that followed NEMA's motor standards would accomplish this. In doing so, any future project or research group that decides to use this project as a starting point will not need to locate the exact motor used here, but only a size 17 NEMA stepper motor.
  • Hybrid Stepper:
Although there are many types of stepper motors, we decided to use the hybrid stepper motor for this project. Aside from the fact that hybrid stepper motors are the most common type, they also have a higher resolution compared to a standard permanent magnet stepper motor. Both wikipedia and NMB's website have some good resources on stepper motor theory and operation.
  • Bipolar:
Although there are several possible winding configurations stepper motors can take, we chose a bipolar motor. This configuration demands the use of an H-bridge in the driving circuit, but allows for higher torque than a unipolar configuration.
  • No Gearhead:
The decision not to use a gearhead was a significant departure from the design of the E-puck. Although a gearhead allows for lower current draw, smaller motors (lighter robot) for any given torque, the gearheads themselves are often composed of fragile plastic components that break easily. Another drawback of geared stepper motors is their higher cost. The benefits and drawbacks of our decision are evident in the final design through its significantly higher weight and improved top speed.
  • Cost:
An average hybrid stepper motor from an established supplier like NMB, Lin Engineering, or Annaheim Automation costs around $40. Although the motors purchased from these companies may provide superior performance, the motors used in this project were purchased from SparkFun for $15 each and provide enough torque to propel our hefty robot at four times the speed of an E-Puck.


SolidWorks assembly of the robot

The focus of the chassis design was ease of manufacture. In order to achieve this, the chassis (and wheels) was designed so that it could be cut out of a single sheet of cast acrylic and simply glued together. The 6mm acrylic used came from McMaster-Carr. In order to minimize the amount of acrylic needed, it was assumed that the circuit board would act as the top of the robot and the front and back were left as short as possible without compromising strength. The final chassis dimensions are W:90mm, L:90mm, H:75mm. As can be seen in the CAD image, the batteries rest on the motors themselves, and, therefore, do not require any additional mounting structure.

  • Wheels:
In order to allow for wheel size adjustments throughout the design and development process, the robot's wheels were made out of acrylic and cut on the laser cutter. The center hole was cut to have a 0.005" clearance so that the motor shaft would fit snugly and act to center the wheel. These shaft collars were super glued to the acrylic wheel and used to fix the wheel to the shaft.
Note: If you are going to cut your own wheels on the laser cutter be mindful that the laser has a tendency to taper the edges of the wheels and does not cut a perfectly round circle due to the resolution of the carriage.
  • Casters
Because the robot has two wheels, it needed supports on both the front and back to prevent it from rocking back and forth or scraping the chassis on the floor. These casters were used in order to reduce friction and allow for simple shim height adjustment.
Note: Although the stainless steel ball in these casters will last longer, they make quite a bit of noise rattling around in their cage. The solution to this would be either to have them perfectly shimmed so they no longer are able to move, or order the equivalent caster with a plastic bearing.

CAD Download

The SolidWorks CAD files can be found here.


Circuit Schematic
Initial breadboard version of the circuit
Final proto-board version of the circuit


The circuit for the robot consists of the NU32 board, two L293D h-bridges (Digi-Key part number 497-2936-5-ND), an XBee wireless chip, connectors for the motors, and a power jack. A circuit diagram of the final version can be found to the right, as well as pictures of the initial breadboard version and final proto-board version.


This project used Lithium Iron Phosphate (LiFePO4) batteries. These are essentially the next generation of Li-Ion batteries. Like Li-Ion cells, they have a high discharging current, but they are also non-explosive when charged improperly. A drawback is that their energy density is slightly less than a good Li-Ion cell. More information about this type of battery can be found here

For this project we used a battery pack that consisted of 4 3.2V A123Systems cells with its own integrated management circuit that prevents the user from charging/discharging the cells past a safe level and regulates the output current above a certain point.

Specifically, we used a 12.8V, 1500mAh, 4A max battery pack that can be purchased here. The charger for this battery pack can also be found here


One of the main goals of the robot was to have the ability to communicate wirelessly. The XBee wireless chip was used to fill this requirement. The XBee is a device that allows simple serial communication over a 2.4 GHz connection. By connecting the chip to ground, power, a transmit pin, and a receive pin, one is able to perform standard serial communication between two properly configured XBee chips. More information about the XBee can be found here and here.


The control software for this project was written for the PIC32 with Microchip's MPLab IDE and C32 compiler. As stated in the original project goals, the code's final form needed to be in the form of a library that a future user can include and easily interface with. As such, the download file below contains three files. One is the main file that does little more than include the project code's header file and call one function in an infinite loop. The second file is a header file that contains the necessary function definitions. The third is the source code that contains all the code written for this project. This includes the interrupt routines, communication, motor control, odometry, and other smaller support functions.

Another goal of the software was to give the user access to three modes of control. These include wheel speed, position in (x,y,θ), and velocity vector control. Wheel speed control allows the user to directly specify each of the wheel speeds at a rate governed by the serial communication speed communication speed. This is the mode of control that the Wiimote code uses. (Also, for reference, it takes the current version of code 133ms of every second to decode these strings if they are being sent at 120Hz.) Position control allows the user to specify the robot's configuration in a global frame relative to its starting position. Velocity vector control allows the user to give the robot a velocity vector (in the global coordinate system) and the center of the front face will track that vector until told to stop.

All of these modes of control support wireless communication through the XBee wireless chip. Furthermore, with 120Hz communication, and all else considered to be the worst-case scenario, this version of the code will only take about 30% of the available computation time.

Code Download

The fully documented code used in this project can be downloaded here, and a flowchart illustrating the functionality of the code can be found here.

Graphical User Interface

Processing window

A GUI was created in Processing in order to demonstrate some of the functionality of the control software and facilitate testing.

An image of the GUI as it appears when it is first opened can be seen on the right.

Modes of Operation

The program has two modes of operation that can be selected with the "mode" button.

Note: The current mode can be discerned by looking at the white/black box next to the mode button and comparing with the legend at the bottom of the GUI

First, there is a position control mode that allows the user to send the robot to a point in a global frame. This is done by clicking in the white box that takes up the majority of the screen space. Once designated, the new position will appear with a number above it which marks its order relative to all the other requested positions. The white area maps to the surface that the robot is sitting on, and the x-axis, which the robot is aligned with, goes to the right of the screen from the origin at the center. The y-axis originates from the same location and goes upward. Both the location of the origin on the screen and the physical dimensions of the space that the screen area maps to are easily adjustable in the code.

Second, there is a velocity control mode that allows you to specify a velocity vector that the center of the front face of the robot must follow. As before, this vector is created by clicking in the white box at the center of the GUI. The robot is, once again aligned along the x-axis, and the magnitude of the velocity vector is specified by the distance from the origin to the point clicked.

Some other features include the following abilies:

  • stop the robot with the "stop" button
  • move the origin to the robot's current position with the "reset" button
  • send the robot back to its starting position with the "home" button
  • specify the robot's default wheel speed for configuration control with the "speed" textbox

Processing Download

You can download the processing project here.

Wii Demonstration

This demo involved allowing a user to drive the robot using a Nintendo Wii remote (Wiimote). The user holds the Wiimote sideways in both hands such that the four arrows of the d-pad are in the left hand. This grip is similar to one used for many Wii games (and a lot like the original Nintendo Entertainment System controller). Holding different combinations of the "1" and "2" buttons then change how fast the robot moves when arrows are also being held. At any point in time, the user can hold the "home" button to send the robot back to its initial configuration. More details on this operation can be found in the WiiMain.c code included below.

The software relies on the wiiuse library. The file linked below includes the important files from the wiiuse package (a header file, a lib file, and a dll file). The Wiimote is connected to a PC (that must be running Windows) via bluetooth, and the wiiuse library is used to interpret the data sent to the PC. Once the data has been interpreted, it is sent to the robot over XBee.

In the file below, there is also an exe file of the WiiMain.c code. In order to execute this file, the wiiuse.dll must be in either the same directory as the executable, or in another directory that is included in Window's search path. To see Window's search path, simply type "echo %PATH%" into a command prompt. Additionally, this executable is assuming that the XBee chip is connected to the computer via "COM3". If a different port must be used, the exe file will need to be re-compiled after editing the pertinent lines in WiiMain.c.

Wii Download

A zip file with the necessary components to run this demo can be found here.

Summary and Future Work

In the end, the robot performed well and met the design requirements. Although it is slightly larger and much heavier than the e-puck, it has similar open-loop dead reckoning abilities, a much higher top speed, and a significantly lower cost (about $180 instead of $970). The first version of the code is reliable and relatively efficient, and the circuit seems to perform as expected and without error. That being said, there are things that could be improved. They are outlined in the following:

  • Wheels:
One of the major factors for any imprecision in our odometry is the relatively low quality wheels that were used. During the prototyping phase of the project, it was very beneficial to have acrylic wheels so that new ones could be cut at any time. Unfortunately, the laser cutter does not do the best job of cutting a perfect circle and also tends to taper the edges of the wheels. Also, while it was convenient to make the wheels out of the same thickness acrylic as the body, thinner wheels would allow the robot to be more precise. Finally, rubber bands were used to increase traction between the wheel and the driving surface, but, after experimenting with the E-Puck, it is clear that far more traction could be generated with a carefully chosen rubber o-ring. All of this being said, it is our recommendation that before any further work is put into making this robot more precise, the wheels be replaced with something similar to those found on an E-Puck
  • Chassis and PCB:
Another significant improvement that could be made would be to reduce the size of the robot. This could be done by designing a PCB specifically for this application, and revising the chassis. If both of these measures were taken, the robot could be made significantly smaller and lighter.
  • Motors:
As far as the weight problem, the main issue is the motor choice. As a proof of concept, the $15 stepper motors were a perfect choice. They had plenty of torque and were very physically robust. As the robot is refined, however, a top priority will likely be to make it lighter. If the motor price is increased to something closer to $40, there are a range of hybrid NEMA stepper motors that are still capable of producing the necessary torque, but are smaller and lighter. Some excellent resources to check are Anaheim Automation, NMB, and Lin Engineering.
  • Code:
There are a few ways the code could be improved. At the moment, it uses two timers to perform its motor control functions. There are ways to do this with only one timer, and implementing these would free up one timer for the user. An example of this can be found in the E-Puck's motor control code. Another improvement would be the addition of some basic functions like forward(speed), reverse(speed), rotate_left(speed), etc... Although the functionality would be identical, changing the current convention of using two SetSpeed() commands to a single function call would both shorten the code and make it more readable. This could also be extended to other tasks performed in the current code that are done frequently.


This project was created by John (Jake) Ware and Jarvis Schultz. Both are mechanical engineering PhD students at Northwestern University.


"Jarvis Schultz" <jarvisschultz2012@u.northwestern.edu>

"John Ware" <johnware2015@u.northwestern.edu>