Do a Google search on “build your own robot” and you’ll find a treasure trove of kits, articles, and videos. Have you ever wanted to give this a try? An engineer at Maxim with a long-standing interest in robotics found a way to build his own using one of the company’s featherboards. Presented with an opportunity to write some example code for the Bosch Sensortec BMI160 inertial measurement unit (IMU) on Maxim’s MAX32630FTHR board, the engineer jumped at the chance to build his own self-balancing robot.
The engineer started his project by configuring the on-board IMU, which estimates a pitch angle from vertical to help the robot stay straight. Should the robot drift away from the vertical point, the control loop would direct the DC motors to keep the base of the robot under its center of gravity. The IMU samples data from the accelerometer/gyroscope and runs it through a complementary filter to produce an estimated feedback signal (providing the estimated pitch of the robot at a given time) for the control loop. The complementary filter, which fuses accelerometer and gyroscope data into a single feedback signal, provides an easier-to-use option than the traditional Kalman filters that can also be used for this purpose.
Performs Floating-Point Calculations Quickly
The MAX32630FTHR board is designed to help engineers quickly implement battery-optimized solutions with the MAX32630 ARM Cortex-M4F microcontroller. This microcontroller features a floating-point unit and performs calculations quickly. With this feature, the engineer didn’t have to perform any approximations to generate the pitch estimates for his robot. The board also features 512K of SRAM, which the engineer used for data logging while the control loop was running. He used, for example, a sample rate of 1.25ms (800Hz) and recorded 14400 samples of the accelerometer horizontal and vertical axis, the gyroscope x-axis, and the pulse-width modulation (PWM + DC) sent to each motor. All of this used only 45% of the available RAM, consisting of 18 seconds of data. Since the board provides a micro SD card connector, the engineer was able to write the saved data to an SD card as a CSV file and then remove the card for data analysis on a PC.
While the engineer could have used various other microcontrollers as the underlying technology for his robot, he chose the MAX32630FTHR board based on all of the features it includes. The 0.9-inch x 2.0-inch board includes a MAX14690 power management IC, a dual-mode Bluetooth module, and a 6-axis accelerometer/gyroscope. It’s ideal for fast proofs-of-concept and early software development of applications like wearable medical patches, portable medical devices, fitness monitors, sports watches, sensor hubs, and, of course, robots.
Figure 1 demonstrates the control loop used for this project.
The forward path of my control loop implements a Proportional, Integral, and Derivative (PID) controller with the output being the duty cycle of the PWM signals driving the MAX14870 H-Bridge for each motor. The PID controller is one of the most widely used controllers in systems today. One of the reasons for its wide use is the ability to ‘tune’ the system by adjusting the PID gains without having a well defined model of the system. This is usually done from an intuitive feel of how changing the PID gains will affect the system and how the system should respond to a disturbance. Procedures such as the Zeigler-Nichols method have been developed to guide that intuition.
For a very good and practical introduction to using a PID controller in a robotic system I recommend the following link: A PID Controller For Lego Mindstorms Robot. The author does an excellent job of introducing each term and the implications of adjusting that term. A variant of the Ziegler-Nichols method of tuning a PID controller is also given.
The feedback path of my control loop uses a Complementary filter. As I mentioned in the introduction, I first learned about this filter in the referenced article. In addition, I found the following references helpful.
1 Reading a IMU Without Kalman: The Complementary Filter
- The Balance Filter
I found the last reference fairly complete in its description of the pros and cons of using each sensor (accelerometer/gyroscope) individually, and why the complementary filter is a suitable solution for ‘fusing’ the two together. The end result is the combination of high pass filtering the integration of angular velocity (gyroscope output), which gives you angular position in that plane, combined with low pass filtering an angle estimate obtained by taking the tangent inverse of the horizontal and vertical components of acceleration in the same plane.