ATtiny Sonar Controller

HC-SR04 sensor

Remember these guys? HC-SR04 ultrasonic rangefinders. Colin used to have three of them, but I’ve recently been reworking his sensor layout. I’m starting by adding five more sensors for a total of eight. Colin’s Arduino has just barely enough free pins to accommodate eight sensors, but I don’t want to spend all of my available pins on sensors. That would leave me no room to expand. Further, I’m not totally sure eight sonar sensors will be enough. My HC-SR04 sensors have a range of about 15 degrees, so it would take 24 sonar sensors to cover 360 degrees. I don’t know if that will be necessary, but I’d like to have the option.

There are a couple of ways to do this. The simplest would probably be to use a shift register. Shift registers allow you to use three I/O pins to control more than a thousand additional pins, but operating them involves some computing overhead and the Arduino still has to be involved in every sensor reading operation.

The chip for my sonar controller, an ATtiny84

An Atmel ATtiny84

The method I’ve chosen is to use a separate microcontroller, an ATtiny84, to handle the sonar sensors. The Aduino tells the ATtiny that it needs new sensor readings. Then it goes and performs some other computations while the ATtiny pings its sensors. Then, when the new sensor data is ready, the ATtiny sends back the new readings all at once. Using this scheme, the Arduino doesn’t have to spend processor time reading sensors. Instead it can focus on other problems. Also, since it can communicate with the Arduino via I2C, it only requires 2 I/O pins! Here’s what Colin looks like with his new sonar layout.

Colin with his new sonar

Colin with his new sonar ring

As with my earlier post on Raspberry Pi to Arduino communication, there’s several steps to this process. Luckily it’s a lot simpler this time.


The Communication Protocol

The communication protocol is going to be very similar to the one that I outlined for communicating between a Raspberry Pi and an Arduino, only simplified. Basically, the Arduino tells the ATtiny when it needs updated sonar readings. The ATtiny then pings all of its sensors and sends back the readings when it’s done. This is how it works:

  • The Arduino sends a byte to the ATtiny
    • The byte has no meaning, it’s just a flag to signal the ATtiny to ping its sensors
  • The ATtiny pings its sensors
  • The ATtiny sends its readings back to the Arduino as 16 bytes
  • The Arduino assembles the bytes into 8 16-bit ints

One other difference with the Raspberry Pi to Arduino protocol is that we’ll be using I2C instead of serial. If you’re not familiar with I2C I’d suggest reading up on it. Sparkfun has a great tutorial.

That’s really all there is to it! The protocol is simple, but there’s some fiddly details to work out in the code.

Top of page


The Sonar Controller

sonar controller with ATtiny84

My ATtiny84-based sonar controller

My sonar controller basically consists of an ATtiny84, with eight inputs for HC-SR04 ultrasonic rangefinders. If you’re interested in making your own, you can download my eagle files for the schematic, board layout, and the gerber files for fabbing the PCB. The gerbers are formatted for fabrication by OSH Park, which I highly recommend for low-volume jobs.

The controllers each have two sets of inputs for power and I2C communication so multiple controllers can be chained together on the same bus. Note that there are two spots for 4.7 kOhm resistors on each controller. These are pull-ups for the I2C buses. If you’re chaining multiple controllers together, only one of them needs pull-up resistors. The resistor spots on the other controllers can be left empty.

I also designed a laser-cut frame to hold the sonar sensors and controller, as well as some fittings to attach the whole business to Colin. The design is pictured below, and you can download the SVG files to make your own parts here.

Colin's new sonar arrangement

Colin with his new sensor arrangement and fittings

If you’d rather not fab a custom circuit board, you can set it up on a breadboard as shown below. It’s not particularly useful on a breadboard, but it will allow you to experiment with and test it.

breadboard wiring

Breadboard wiring for ATtiny sonar controller

Top of page


The Code!

ATtiny Code

Let’s start out with a couple of preliminaries. First, for I2C communication I’m using the TinyWireS and TinyWireM libraries. We need our ATtiny to perform as both master and slave, but there is no existing library that allows this. Fortunately there’s a way to hack around this problem.

I’m also using the NewPing library. The NewPing library doesn’t work with ATtinies, but there’s a hack for this too. You just need to comment out the functions for Timer2 and Timer4 because the ATtiny does not have these timers.

You can find the complete program for the ATtiny here. If you’re not familiar with how to program ATtinies, I’d suggest going through this very thorough tutorial.

Initiating A Sensor Update

The function below pings the eight sonar sensors and records the ping times in microseconds.

void updatePingTimes(uint8_t bytes)
{
  TinyWireS.receive();
  for (int i = 0; i < NUM_SONAR; i++)
  {
    pingTimes[i] = sonar[i].ping();
  }
  sendTimes();
}

The function above is invoked using an interrupt when the ATtiny receives a byte from the Arduino. The line below is called in the setup() function to tell the ATtiny to generate an interrupt and call the updatePingTimes() function whenever data is received from the Arduino.

TinyWireS.onReceive(updatePingTimes);

Sending Data Back To The Arduino

When the sonar sensors have been updated, the sendTimes() function is called to send the updated ping times back to the Arduino. It is important to note that the ATtiny must be functioning as a slave in order to use the onReceive() function. To send data back to the Arduino, the ATtiny must function as a master. This makes the sendTimes() function a bit more complicated.

void sendTimes()
{
  // clear the I2C registers
  USICR = 0;
  USISR = 0;
  USIDR = 0;

  // start I2C with ATtiny as master
  TinyWireM.begin();

  // transmit ping times to Arduino
  TinyWireM.beginTransmission(MASTER_ADDRESS);
  for (int i = 0; i < NUM_SONAR; i++)
  {
    int thisTime = pingTimes[i];
    if (thisTime == 0) thisTime = (double)MAX_DISTANCE / speedOfSound;
    uint8_t firstByte = thisTime & 0xFF;
    uint8_t secondByte = (thisTime >> 8) & 0xFF;
    TinyWireM.write(firstByte);
    TinyWireM.write(secondByte);
  }
  TinyWireM.endTransmission();

  // clear I2C registers again
  USICR = 0;
  USISR = 0;
  USIDR = 0;

  // put ATtiny back in slave mode
  TinyWireS.begin(OWN_ADDRESS);
}

As I said before, I don’t know of any library that allows an ATtiny to function as both master and slave. Fortunately we can hack our way around this problem by clearing the ATtiny’s I2C registers and restarting communication in master mode. After we’ve transmitted the ping times back to the Arduino we need to clear the I2C registers again and put the ATtiny back in slave mode so it can be ready for the next request from the Arduino. I don’t mind telling you it took me a lot of time with the ATtiny’s datasheet to figure out how to make that work.


Arduino Code

For reference you can find the complete program for the Arduino here. To get things going we need to include the Wire library and put the following two lines in the setup() function:

// begin communication with sonar controller
Wire.begin(OWN_ADDRESS);
Wire.onReceive(updateDistances);

Adding the onReceive() function is particularly important because it allows the Arduino to do other processing tasks while the ATtiny pings its sensors. When the ATtiny sends data it generates an interrupt, the Arduino stops what it’s currently doing and receives the new data, then goes back to whatever it was doing before.

Receiving Data

The updateDistances() function works as follows:

// updates sonar distance measurements when a new reading is available
void updateDistances(int bytes)
{
  for (int i = 0; i < NUM_SONAR; i++)
    readSonar(i);
  distancesRead = true;
}

// reads the distance for a single sonar sensor
// expects to receive values as ints broken into 2 byte pairs,
// least significant byte first
void readSonar(int index)
{
  int firstByte = Wire.read();
  int secondByte = Wire.read();
  sonarDistances[index] = ((double)((secondByte << 8) | firstByte)) * speedOfSound * 0.5;
}

Notice that the sonar distance is multiplied by 0.5 when it’s added to the sonarDistances array. This is because the ATtiny returns the total time of flight for the ultrasonic ping. The ping needs to go from the sensor, to the obstacle, and back. This means multiplying the ping time by the speed of sound would result in twice the distance between the obstacle and the sensor.

Requesting An Update

Fortunately, requesting an update is pretty simple. The Arduino just needs to send a byte, any byte, to the ATtiny to let it know it needs new sensor readings.

// requests a sonar update from the sonar controller at the given address
// by sending a meaningless value via I2C to trigger an update
void requestSonarUpdate(int address)
{
  Wire.beginTransmission(address);
  Wire.write(trig);
  Wire.endTransmission();
}

After this function executes, the Arduino can perform other processing tasks until the ATtiny sends back updated sonar readings.

Top of page


Going Further

There are a number of ways to solve the problem of coordinating large numbers of sonar sensors, and the above method is only one possibility. It takes a long time to update sonar sensors, up to 15 milliseconds for 8 sensors. For an Arduino running at 16 MHz, that’s 240,000 processor cycles. So it’s advantageous to use a method that allows the Arduino to do something else while the sensors are being updated. My method does this, but one could also take the Arduino out of the loop entirely and have the Raspberry Pi talk to the sonar controller directly. It would be interesting to implement this method and compare it to the one presented above.

Top of page

Raspberry Pi to Arduino Communication For Robot Control


Arduino to Raspberry Pi communicationGood news, everyone! I’ve come up with a good way for Colin’s Raspberry Pi to talk to his Arduino. To review, the idea was that the Arduino could handle low-level control functions like speed control, odometry, and sensor reading. This leaves the Raspberry Pi free to handle high-level control like obstacle avoidance, motion planning, and state estimation.

There are a few steps in this process:


Freeing Up the Raspberry Pi’s Serial Port

The first problem we have to deal with is that the Raspbian reserves its serial port for use by a serial console. So the Raspberry Pi’s GPIO serial port is totally useless until you free it up. I’m not going into all the details here, but I found this guide really helpful. The important things to remember are that you need to enable uart in config.txt and disable the serial console. This will allow our program to use the serial port.

Top of page


The Communication Protocol

The communication protocol works like this:

  • The Raspberry Pi sends 2 16-bit ints to the Arduino
    • The first int is the commanded translational velocity
    • The second int is the commanded angular velocity
  • The Arduino sets its speeds accordingly and then updates its sonar sensors
  • After the sensors are updated, the Arduino sends 11 16-bit ints back to the Raspberry Pi
    • The first 8 ints are the distance readings from the 8 sonar sensors
    • The last 3 ints are the Colin’s x and y position and his heading, calculated from odometry

Using this protocol, Colin will run at the commanded speeds until he receives another command. This causes a problem: Colin will continue to run even if the serial communication fails and the Raspberry Pi stops sending commands all together. This means he could run off a cliff and there would be nothing to stop him!

To fix this I have the Raspberry Pi send commands at a regular interval. If, for example, Colin expects to get a new command every quarter second and he doesn’t get a command at the expected time, he will know there’s a communication problem. If Colin detects a communication fault he can respond in a fail-safe manner by stopping.

Communication Formatting

How do we send ints over serial though? Arduino’s Serial.print()  function converts int and float values to strings of chars before sending them. At first this might seem really convenient, but it actually causes more problems than it solves. First, Arduino and C++ don’t have a great function for converting char strings to int or float values. Second, C++ doesn’t automatically convert numerical types to strings before sending them via serial, so you’d need to write your own function for that. Lastly, if we convert numbers to strings, their lengths will be variable. This means we would need to define some way to tell when one value ends and the next begins.

The good news is none of that is necessary! You know why? Every command and sensor packet will be exactly the same length: 2 16-bit ints for commands and 11 16-bit ints for responses. Further, the ints will always come in the same order. So the meaning of each byte is predictable.

The only problem is you can only send individual bytes over serial. But this is easily solved by splitting the ints into their component bytes before sending:

char firstByte = (byte)(value & 0xFF);
char secondByte = (byte)((value >> 8) & 0xFF);

and reassembling them on the other end:

int value = (secondByte << 8) | firstByte;

Top of page


Wiring It Up

Wiring up the Raspberry Pi to the Arduino is pretty simple, but there’s an important catch. The Raspberry Pi uses 3.3 volt logic and the Arduino uses 5 volt logic. So we need to use a level shifter to allow communication between the two devices. If the level shifter gets a 3.3 volt signal on the low side, it sends out a 5 volt signal on the high side. If it gets a 5 volt signal on the high side it sends out a 3.3 volt signal on the low side. Pretty simple, right? Wire it up as shown below:

wiring for serial between an rPi and an Arduino

Wiring for serial between a Raspberry Pi and an Arduino


Top of page


The Code!


Okay, enough talk. Let’s get into the code. I’m going present the code for the Raspberry Pi side first, and follow it up with the Arduino code. The complete code for the Raspberry Pi can be found here and the Arduino code can be found here.

Raspberry Pi Code

Opening the Serial connection

First, we need to open a serial connection with the Arduino. This is handled by the following function, which I adapted from this extremely helpful site. Check out that site if you want details on how all of this works.

void SerialBot::openSerial()
{
	serialFd_ = open("/dev/serial0", O_RDWR);
																						// to allow blocking read
	if (serialFd_ == -1)
	{
		cerr << "Error - unable to open uart" << endl;
		exit(-1);
	}	
	
	struct termios options;
	tcgetattr(serialFd_, &options);
	options.c_cflag = B9600 | CS8 | CLOCAL | CREAD;
	options.c_iflag = IGNPAR;
	options.c_oflag = 0;
	options.c_lflag = 0;
	tcflush(serialFd_, TCIFLUSH);
	tcsetattr(serialFd_, TCSANOW, &options);
}

There is one important difference between my function above and the one it’s adapted from: I dropped the O_NOCTTY and O_NDELAY flags from the open command in line 3. This means my serial connection will be blocking. In other words, when I call the read() function the program execution stops until there is data to read in the serial buffer. In other words, my program will wait for a response from the Arduino before continuing.

Sending Commands

Sending a command works as follows:

int SerialBot::transmit(char* commandPacket)
{
	int result = -1;
	if (serialFd_ != -1) 
	{
		result = write(serialFd_, commandPacket, commandPacketSize);
	}
	return result;
}

And, in case you’re wondering, the function that assembles the commandPacket is below.

void SerialBot::makeCommandPacket(char* commandPacket)
{
	int16_t intAngular = (int)(angular_ * 1000.0);
	commandPacket[0] = (char)(translational_ & 0xFF);
	commandPacket[1] = (char)((translational_ >> 8) & 0xFF);
	commandPacket[2] = (char)(intAngular & 0xFF);
	commandPacket[3] = (char)((intAngular >> 8) & 0xFF);
}

Note that angular_ is a double representing Colin’s commanded angular velocity. The size and representation of doubles is inconsistent, however, so it’s difficult to break them up and reassemble them on a different machine. Int representations are very consistent, however, so I just multiply angular_ by 1000 to save the first three decimal places and cast it to an int. The loss of accuracy is pretty negligible for our purposes.

Receiving Data

The function below is how we receive data from the Arduino. Note that I’ve set the read to time out after 0.25 seconds. The Raspberry Pi expects to get a response from the Arduino after every command is sent. If it doesn’t receive a response before it’s time to send the next command, it throws an error.

int SerialBot::receive(char* sensorPacket)
{
	memset(sensorPacket, '\0', sensorPacketSize_);
	int rxBytes;
	if (serialFd_ != -1)
	{
		// set up blocking read with timeout at .25 seconds
		fd_set set;
		FD_ZERO(&set); // clear the file descriptor set
		FD_SET(serialFd_, &set); // add serial file descriptor to the set
		struct timeval timeout;
		timeout.tv_sec = 0;
		timeout.tv_usec = 250000;
		
		// wait for serial to become available
		int selectResult = select(serialFd_ + 1, &set, NULL, NULL, &timeout);
		if (selectResult < 0)
		{
			cerr << "blocking read failed" << endl;
			return -1;
		}
		else if (selectResult == 0)
		{
			cerr << "read failed: timeout occurred" << endl;
			return 0;
		}
		else
		{
			rxBytes = read(serialFd_, sensorPacket, numSonar_ + numPoseVariables);
		}
	}
	return rxBytes;
}

Once we’ve read data from the Arduino, we need to parse it:

int SerialBot::parseSensorPacket(char* sensorPacket)
{
	int16_t firstByte;
	int16_t secondByte;
	int16_t inValues[numSonar_ + numPoseVariables];
	for (int i = 0; i < numSonar_ + numPoseVariables; i++)
	{
		firstByte = sensorPacket[2 * i];
		secondByte = sensorPacket[(2 * i) + 1];
		inValues[i] = (secondByte << 8) | firstByte;
	}

	for (int i = 0; i < numSonar_; i++)
	{
		distances_[i] = inValues[i];
	}
	
	x_ = inValues[8];
	y_ = inValues[9];
	theta_ = ((double)inValues[10]) / 1000.0;
}

Note again that Colin’s heading, theta_ , is a double. To save some bother in programming, the double value is multiplied by 1000 and casted to an int before it’s sent. So it needs to be casted to a double and divided by 1000 after it’s received.

Putting It All Together

Okay, last thing: we’ll put all of these things together in a communication function that runs every 0.25 seconds in its own thread:

void SerialBot::commThreadFunction()
{
	while (true) 
	{
		char commandPacket[commandPacketSize];
		makeCommandPacket(commandPacket);
		if (transmit(commandPacket) < 1)
			cerr << "command packet transmission failed" << endl;
		char sensorPacket[sensorPacketSize_];
		memset(sensorPacket, '\0', sensorPacketSize_);
		int receiveResult = receive(sensorPacket);
		if (receiveResult < 1)
		{
			cerr << "sensor packet not received" << endl;
		}
		else if (receiveResult < commandPacketSize)
		{
			cerr << "incomplete sensor packet received" << endl;
		}
		else
		{
			parseSensorPacket(sensorPacket);
		}
		usleep(readPeriod_);
	}
}

Top of page


The Arduino Code

Are you still with me? That took a while, but we got one side of it done. So we just have the Arduino code left to deal with.

Receiving

Let’s start with receiving a command from the Raspberry Pi:

void readCommandPacket()
{
  byte buffer[4];
  int result = Serial.readBytes((char*)buffer, 4);

  if (result == 4) // if the correct number of bytes has been received
  {
    int commands[2];
    
    // assemble 16 bit ints from the received bytes in the buffer
    for (int i = 0; i < 2; i++)
    {
      int firstByte = buffer[2 * i];
      int secondByte = buffer[(2 * i) + 1];
      commands[i] = (secondByte << 8) | firstByte;
    }
    translational = commands[0]; 
    angular = (double)commands[1] / 1000.0; // convert received int to double angular velocity
    colin.drive(translational, angular); // set Colin's speeds
    commandReceived = true; // note that a command has been received
    lastCommandTime = millis();
  }
  else if (result > 0)
  {
    Serial.println("incomplete command");
  }
  // else do nothing and try again later
}

Note that I’m using the Serial.readBytes() function rather than the more common Serial.read() function. There’s a couple of reasons for this. First, Serial.read() only reads a char at a time, but we know we need 4 bytes. Serial.readBytes() also blocks the program’s execution until it receives the requested number of bytes. This is perfect, since it means we’ll get a complete packet, instead of just receiving part of one.

Transmitting

The transmit function first puts all the data that needs to be sent into an array, buffer. Then the buffer is sent to the Raspberry Pi using Serial.write() . Note that I’m not using Serial.print() because it automatically converts int values to characters, and we want to send the bytes exactly as-is.

void sendSensorPacket()
{
  colin.getPosition(x, y, theta); // updates Colin's position
  byte buffer[22];
  addDistances(buffer); // adds sonar readings to buffer
  int sendX = (int)x;
  int sendY = (int)y;
  int sendTheta = (int)(theta * 1000.0);
  buffer[16] = (byte)(sendX & 0xFF);
  buffer[17] = (byte)((sendX >> 8) & 0xFF);
  buffer[18] = (byte)(sendY & 0xFF);
  buffer[19] = (byte)((sendY >> 8) & 0xFF);
  buffer[20] = (byte)(sendTheta & 0xFF);
  buffer[21] = (byte)((sendTheta >> 8) & 0xFF);
  Serial.write(buffer, 22);
}

void addDistances(byte* buffer)
{
  for (int i = 0; i < NUM_SONAR; i++)
  {
    buffer[2 * i] = (byte)(sonarDistances[i] & 0xFF);
    buffer[(2 * i) + 1] = (byte)((sonarDistances[i] >> 8) & 0xFF);
  }
}

Bringing It All Together

The loop() function below brings everything together. It checks to see if there is data in the serial buffer and, if so, attempts to interpret it as a command. If it successfully reads a command, it requests an update from the sonar controller. After the Arduino gets updated sensor readings it assembles a response packet and sends it back to the Raspberry Pi.

Lastly, the Arduino checks to see if more than 1 second has passed since the last command was received. If so, it assumes that a communication fault has occurred and it stops Colin.

void loop() 
{ 
  // check if a command packet is available to read
  readCommandPacket();
  
  // request a sensor update if a command has been received
  if (commandReceived)
  {
    commandReceived = false;
    requestSonarUpdate(SONAR_ADDRESS);
  }
  // send sensor packet if sonar has finished updating
  if (distancesRead)
  {
    distancesRead = false; 
    sendSensorPacket();
  }
  currentTime = millis();
  
  // stop colin if a command packet has not been received for 1 second
  if (currentTime - lastCommandTime > 1000)
  {
    Serial.println("command not received for 1 second");
    lastCommandTime = millis();
    colin.drive(0, 0.0);
  }
}

Top of page


Where Do We Go From Here?

Now we have a good way to communicate between Colin’s Raspberry Pi and Arduino. Colin doesn’t have a way to perceive the world around him, however, so that’s our next step. I designed an independent controller to read Colin’s sonar sensors and relay the information to the Arduino via I2C. My next post will cover the finer details of my sonar controller and the associated communication protocol.

After we sort these details out and get the system working like we want, we can get to programming higher level behaviors. For example, I’m currently working on a wall-following program. I’m hoping it will be ready to present in a couple of weeks! Lots of good stuff to come, stay tuned!

Top of page

Odometry With Arduino

Now that we can control the speed of Colin’s wheels, we can tell how far Colin has moved using odometry. It involves counting the encoder ticks for Colin’s motors and integrating that information over time to determine Colin’s change in position. This method has the distinct advantage that it relies on the actual motion of Colin’s wheels, and thus doesn’t require absolute accuracy from the speed control algorithm. Odometry also provides a good motion model that can be used as part of a larger localization algorithm. As such it’s a good stepping stone toward my goal of making a simultaneous localization and mapping program for Colin!

This tutorial owes a lot to MIT’s primer on odometry and motor control. It does a great job explaining the theory behind odometry.


Theory Basics

The position of a robot in space is referred to as its pose, which is defined by six quantities: its translation in Cartesian coordinates (x, y, and z) and its rotation about those three axes (θx, θy, and θz). Luckily, a differential drive robot like Colin can only translate in two dimensions and rotate in one, so Colin’s pose can be defined by three quantities (x, y, and θz).

Let’s say Colin’s initial pose is (0, 0, 0) at t=t_{0}. How can we determine his change in pose when t=t_{0}+\Delta t where \Delta t is the time interval between pose updates? Because we’re already using encoders to control Colin’s speed, it’s easy to keep track of the distance Colin’s wheels have turned. In fact, my Encoder class already does this with its getDistance() function.

Let’s say d_{left} is the distance turned by the left wheel over \Delta t, and d_{right} is the same quantity for the right wheel. Knowing these two distances can tell us a couple things. If d_{left}=d_{right} then Colin traveled in a straight line. If d_{left}\gt d_{right} he turned to the right and if d_{left}\lt d_{right} he turned left. We can also use d_{left} and d_{right} to calculate Colin’s exact translation and rotation.

To simplify things a bit we’ll assume Colin’s wheel speeds are constant, which adds a negligible amount of error as long as we keep \Delta t small. This assumption means that Colin is always travelling along a circular arc. The length of this arc, d_{center} is given by the average of d_{left} and d_{right}:

d_{center}=\frac{d_{left}+d_{right}}{2}

We’ll say that Colin’s rotation in radians over \Delta t is \phi. Also, let r_{left} be the distance between the center of Colin’s arc of travel and his left wheel and r_{right} be the same distance for the right wheel. This means that d_{left}=\phi r_{left} and d_{right}=\phi r_{right}. Also, r_{left}=r_{right}+d_{wheels} where d_{wheels} is the distance between Colin’s wheels. With a little bit of algebra we can show the following:

\phi=\frac{d_{right}-d_{left}}{{d_{wheels}}}

We can also calculate Colin’s change in his x and y coordinates via the following equations:

x'=x+d_{center}cos(\theta)

y'=y+d_{center}sin(\theta)

Where x' and y' are the new x and y position, respectively. It’s important to note that the above equations are simplified. They assume that Colin’s motion happens in two discrete phases: he rotates in place and then translates along a straight line. This is clearly not true, but as long as \phi is small, the error introduced is negligible. This means that, as with our prior simplification, we need to keep \Delta t small to make this work. I’m not going to go into all the details here, but if you’re interested you can find the full derivation in the MIT odometry tutorial.

So, now that we have worked out the mathematical underpinnings for odometry, we can translate this into code!


Odometry Code

The magic happens in my new DifferentialDrive library. We’ll just go over the odometry portion today, but DifferentialDrive allows the user to control an arbitrary differential drive robot by specifying the robot’s translational and angular velocities and, optionally, the distance the robot should travel. I’ll explain all of that in a later post and include some implementation examples as well!

void DifferentialDrive::updatePosition()
{
   // get the angular distance traveled by each wheel since the last update
   double leftDegrees = _leftWheel->getDistance();
   double rightDegrees = _rightWheel->getDistance();

   // convert the angular distances to linear distances
   double dLeft = leftDegrees / _degreesPerMillimeter;
   double dRight = rightDegrees / _degreesPerMillimeter;

   // calculate the length of the arc traveled by Colin
   double dCenter = (dLeft + dRight) / 2.0;

   // calculate Colin's change in angle
   double phi = (dRight - dLeft) / (double)_wheelDistance;
   // add the change in angle to the previous angle
   _theta += phi;
   // constrain _theta to the range 0 to 2 pi
   if (_theta > 2.0 * pi) _theta -= 2.0 * pi;
   if (_theta < 0.0) _theta += 2.0 * pi;

   // update Colin's x and y coordinates
   _xPosition += dCenter * cos(_theta);
   _yPosition += dCenter * sin(_theta);
}

The above function needs to be called every \Delta t and, to keep the error from our simplifications small \Delta t needs to be small. In my testing I’ve found that doing a position update with the same frequency as the updates for the PID motor controller (every 50ms) results in good accuracy over short distances. However, this update involves a significant amount of extra computation, and doing it 20 times per second might require an excessive amount of processor time if you’re trying to do a lot of other computation at the same time. I’ve found that doing position updates half as often (every 100ms) results in very little loss of accuracy, so it’s entirely possible to balance accuracy and the resources your program has to spare.


Further Work

First of all, we need to integrate the above update function into the larger class that controls Colin’s motion. I’ll demonstrate that in a later post and include some examples that show how to use the class in an Arduino sketch.

Also, odometry can only be used to calculate Colin’s position relative to his starting position. It cannot be used to determine his absolute position in a space unless his starting position is known.

The larger problem is that odometry is inherently inaccurate. Encoder ticks do not translate directly into distance traveled by the wheel because wheels slip, the wheels aren’t perfectly circular, the ground isn’t perfectly flat, encoder ticks might be missed, and the motor gearbox has backlash that isn’t accounted for in our model. This means that Colin’s position calculated from odometry will gradually diverge from his true position. We could use other methods that might be more accurate, such as optical flow and IMUs. However, any sensor we might use suffers from some inherent random error, known as noise, and this error will accumulate over time.

To compensate for this error we can calculate Colin’s probable position by incorporating data from another sensor. This is what I’ll be working on over the next several months. First I’ll develop a program to localize him to a pre-existing (or a priori) map, and then I’ll work on a program that allows him to build his map on the fly.

I should note that software for this purpose is already available as part of the robot operating system (ROS), but I’m not interested in pre-made solutions. My goal here is to develop these solutions myself so we can all learn the intimate details of their operation.

Motor Encoders with Arduino

If Colin is going to create a map of the space he’s in, he needs to be able to find the distance between objects in the space. If he’s going to accurately control his own motion, he needs to be able to control the speed of his motors. For both of these things he needs information about how far and how fast his motors are turning. To give Colin that information I added encoders to his motors. Encoders are special sensors that track both how far his motor shafts have turned, and in what direction.

The following tutorial will discuss how to use shaft encoders with DC motors. First I’ll discuss the principles the encoders work on, then I’ll show you how to wire up an encoder and verify that it’s working. This tutorial owes a large dept to the Penn State Robotics Club Wiki, which has an excellent page on wheel encoders.

For this tutorial I’ll be using these magnetic encoders from Pololu. They are great for use with an Arduino because their output is a nice, easy to read, digital square wave. Pololu also offers an analog optical encoder, but to reliably read its output you need to use a ADC (analog to digital converter). Also, the optical encoder is susceptible to interference from bright light. So magnetic encoders are easier to use and more reliable for our purposes.


Encoder Principles

Pololu’s shaft encoders add a 6 pole magnetic disc to the shaft of the motor, as well as two Hall effect sensors. As the motor turns the disc rotates past the sensors. Each time a magnetic pole passes a sensor, the encoder outputs a digital pulse, also called a “tick.” The encoder setup is pictured below.

motor encoder installation

motor encoder installation example (from pololu.com)

The encoder has two outputs, one for each Hall effect sensor. The sensors are separated by 90 degrees. This means the square wave outputs of the sensors are 90 degrees out of phase. This is called a quadrature output. The picture below (taken from Pololu’s website) shows the typical output of an encoder.

encoder output

Output of an encoder (from pololu.com)

Why is it important that the output pulses are 90 degrees out of phase? This allows us to determine both the magnitude and the direction of the motor’s rotation. If output A is ahead of output B, then the motor is turning forward. If output A is behind B, the motor is turning backward. Pretty cool, right?


Encoder Setup and Wiring

Wiring up the encoders is pretty simple. See the diagram below for details on wiring up the encoder for motor B, and repeat for motor A.

wiring for a single motor and encoder

wiring for a single motor and encoder

Note that the encoder pin OUTA needs to be connected to a hardware interrupt pin (digital pin 2 or 3 on an Arduino Duemilanove or Uno). For this example it’s not really necessary to use hardware interrupts but if you don’t your Arduino won’t be able to do anything but track the encoder. Interrupts allow your Arduino to keep track of the encoders while it’s running another program.

It’s also important when installing your motors to not mount them too close to each other. If your motors are mounted back-to-back, the magnetic encoder wheels will interfere with each other. I initially had only about 5mm between the two encoder wheels. The magnets in the encoder wheels linked with each other and made it impossible for the two motors to turn at different speeds. Keeping the encoder wheels at least 20mm apart seems to avoid any interference problems.


Hardware Interrupts

It’s pretty easy to write a program that simply counts the number of ticks. But if, for some reason, you want your program to do anything but track the encoders you need to use hardware interrupts.

With hardware interrupts your main program runs until the state of one of the Arduino’s interrupt pins changes. Then the main program is stopped, a special interrupt method is run, then the main program is resumed. For example, an obstacle avoidance routine can run in the foreground but if one of the interrupt pins changes from low to high the position or speed of the motor can be updated in the interrupt method.

Because the main program is stopped when the interrupt method is triggered, the length of the interrupt method should be kept as short as possible. If the interrupt method is too long, then the Arduino will spend all of its time running the interrupt method and it will never actually run the main program.

For more details on hardware interrupts check out this page on the Arduino playground.


Example Sketch

The sketch below simply prints the values of the two count variables. When the motors are moved, the encoder output triggers the appropriate encoder event method. The encoder event methods either increment or decrement the count variables, keeping track of the number of ticks from each encoder. If the motor is turned forward, the count is incremented and it is decremented if the motor is turned backward.

The motors need to be moved manually for this sketch. They won’t move themselves. We will expand this to do simple motor speed control in a later tutorial.

We’ll define motor direction somewhat arbitrarily: If RH_ENCODER_A is HIGH and RH_ENCODER_B is LOW then the motor is turning forward. If RH_ENCODER_A is HIGH and RH_ENCODER_B is also HIGH, then the motor is turning backward. You can swap the wires for ENCODER_A and ENCODER_B to change the direction if required.

/*
 * Encoder example sketch
 * by Andrew Kramer
 * 1/1/2016
 *
 * Records encoder ticks for each wheel
 * and prints the number of ticks for
 * each encoder every 500ms
 *
 */

// pins for the encoder inputs
#define RH_ENCODER_A 3 
#define RH_ENCODER_B 5
#define LH_ENCODER_A 2
#define LH_ENCODER_B 4

// variables to store the number of encoder pulses
// for each motor
volatile unsigned long leftCount = 0;
volatile unsigned long rightCount = 0;

void setup() {
  pinMode(LH_ENCODER_A, INPUT);
  pinMode(LH_ENCODER_B, INPUT);
  pinMode(RH_ENCODER_A, INPUT);
  pinMode(RH_ENCODER_B, INPUT);
  
  // initialize hardware interrupts
  attachInterrupt(0, leftEncoderEvent, CHANGE);
  attachInterrupt(1, rightEncoderEvent, CHANGE);
  
  Serial.begin(9600);
}

void loop() {
  Serial.print("Right Count: ");
  Serial.println(rightCount);
  Serial.print("Left Count: ");
  Serial.println(leftCount);
  Serial.println();
  delay(500);
}

// encoder event for the interrupt call
void leftEncoderEvent() {
  if (digitalRead(LH_ENCODER_A) == HIGH) {
    if (digitalRead(LH_ENCODER_B) == LOW) {
      leftCount++;
    } else {
      leftCount--;
    }
  } else {
    if (digitalRead(LH_ENCODER_B) == LOW) {
      leftCount--;
    } else {
      leftCount++;
    }
  }
}

// encoder event for the interrupt call
void rightEncoderEvent() {
  if (digitalRead(RH_ENCODER_A) == HIGH) {
    if (digitalRead(RH_ENCODER_B) == LOW) {
      rightCount++;
    } else {
      rightCount--;
    }
  } else {
    if (digitalRead(RH_ENCODER_B) == LOW) {
      rightCount--;
    } else {
      rightCount++;
    }
  }
}

That’s all for today. In my next post I’ll take what we learned about using encoders with hardware interrupts and incorporate that into a simple speed control routine. After that we can dive into real PID control!

Bitwise Obstacle Avoidance Tweak

Around the time I was rewriting the example sketch for the obstacle avoidance tutorial I stumbled across this great tutorial on bit math on the Arduino Playground. While reading it I found a good, simple use for bit math that would make my obstacle avoidance sketch a bit more efficient.

There is a popular quote, usually attributed to Don Knuth: “premature optimization is the root of all evil.” Premature optimization occurs when we make improvements to efficiency that complicate the program, reduce readability, or cause us to ignore larger, more structural improvement possibilities.

The following is definitely a premature optimization of the obstacle avoidance example sketch. However, I think it’s okay in this case because it allows us to practice a fun new technique: bit math!


Background

First thing’s first: if you haven’t read the Arduino Playground tutorial on bit math, read it now. It’s good, it’s easy to read, and it provides much more background than I’m going to.

In my obstacle avoidance example I used a multi dimensional array of bools to store the sensor states for each possible reaction, the reaction matrix. Each bool value occupies one byte, so the size of the reaction matrix in bytes is the number of sensors times the number of reactions. In our case it’s 24 bytes.

Each byte is made of 8 bits, each of which have two states. A bool value only has two states, so in theory a bool could be represented by a single bit. Further, a byte could be interpreted as an array of 8 bools. Cool, right? This means, if the number of sensors is 8 or less, the size of our reaction matrix in bytes is just the number of reactions. The size of our reaction matrix is only 8 bytes now!

Additionally, our compareCases()  can be totally eliminated by this. It formerly involved two nested for loops to compare the values in the sensor array with the reaction matrix. The number of comparisons was the equal to the number of sensors times the number of reactions. Now all we have to do is compare the thisCase  byte directly to each possible case in the switch statement. Total number of comparisons is equal to the number of cases. We just cut the number of comparisons for each sensor update down from 24 to 8! This happens ten times per second, so it’s actually kind of significant.


Example Sketch

/*
 * Obstacle avoidance bit math example
 * by Andrew Kramer
 * 8/3/2015
 * 
 * Uses 3 ultrasonic sensors to determine 
 * if nearest obstacle is too close.
 * Outlines 8 possible cases for 
 * positions of obstacles.
 */

#include <NewPing.h>

#define RH_PWM 6 // PWM pin for right hand motor
#define RH_DIR1 9 // direction control for right hand motor
                  // BIN1 pin on motor controller
#define RH_DIR2 8 // direction control for right hand motor
                    // BIN2 pin on motor controller

#define LH_PWM 11 // PWM pin for left hand motor
#define LH_DIR1 13 // direction control for right hand motor
                     // AIN1 pin on motor controller
#define LH_DIR2 12 // direction control for left hand motor
                     // AIN2 pin on motor controller
                     
#define DEFAULT_SPEED 30 // default PWM level for the motors
#define TURN_DISTANCE 20 // distance at which the bot will turn
#define MAX_DISTANCE 200 // max range of sonar sensors
#define NUM_SONAR 3 // number of sonar sensors
#define NUM_CASES 8 // number of reaction cases

NewPing sonar[NUM_SONAR] = { // array of sonar sensor objects
  NewPing(A2, A2, MAX_DISTANCE), // right
  NewPing(A1, A1, MAX_DISTANCE), // front
  NewPing(A0, A0, MAX_DISTANCE) // left
};

/* list of cases
    B0000000  0: no obstacles
    B0000001  1: obstacle to the right
    B0000010  2: obstacle in front
    B0000011  3: obstacles front and right
    B0000100  4: obstacle to the left
    B0000101  5: obstacles left and right
    B0000110  6: obstacles front and left
    B0000111  7: obstacles front, left, and right
*/

byte thisCase = 0;

void setup() {
  for (int pin = 6; pin <= 13; pin++) {
    pinMode(pin, OUTPUT); // set pins 3 through 9 to OUTPUT
  }
  Serial.begin(9600);
}

void loop() {
  updateSensor();
  switch (thisCase) {
    // no obstacles
    case B00000000:
      straightForward();
      break;
    // obstacle to the right
    case B00000001:
      turnLeft(30);
      break;
    // obstacle in front
    case B00000010:
      turnLeft(90);
      break;
    // obstacles front and right
    case B00000011:
      turnLeft(90);
      break;
    // obstacle to the left
    case B00000100:
      turnRight(30);
      break;
    // obstacles left and right
    case B00000101:
      turnLeft(180);
      break;
    // obstacles front and left
    case B00000110:
      turnRight(90);
      break;
    // obstacles front, left, and right
    case B00000111:
      turnLeft(180);
      break;
  }
  delay(100); 
}

void updateSensor() {
    thisCase = 0;
    for (int i = 0; i < NUM_SONAR; i++) {
        int distance = sonar[i].ping_cm();
        if (distance == 0) 
            distance = MAX_DISTANCE;
        if (distance < TURN_DISTANCE)
            thisCase |= (1 << i);
    }
}

void setLeftForward() {
  digitalWrite(LH_DIR1, HIGH);
  digitalWrite(LH_DIR2, LOW);
}

void setRightForward() {
  digitalWrite(RH_DIR1, HIGH);
  digitalWrite(RH_DIR2, LOW);
}

void setBothForward() {
  setLeftForward();
  setRightForward();
}

void setLeftBackward() {
  digitalWrite(LH_DIR1, LOW);
  digitalWrite(LH_DIR2, HIGH);
}

void setRightBackward() {
  digitalWrite(RH_DIR1, LOW);
  digitalWrite(RH_DIR2, HIGH);
}

void setBothBackward() {
  setRightBackward();
  setLeftBackward();
}

void setLeftSpeed(int speed) {
  analogWrite(LH_PWM, speed);
}

void setRightSpeed(int speed) {
  analogWrite(RH_PWM, speed);
}

void setBothSpeeds(int speed) {
  setLeftSpeed(speed);
  setRightSpeed(speed);
}

// sets direction of both motors to forward and sets both speeds
// to default speed
void straightForward() {
  setBothForward();
  setBothSpeeds(DEFAULT_SPEED);
}

// makes a turn by stopping one motor
// accepts an int, 0 or 1 for left or right turn respectively
void turnLeft(int deg) {
  setBothSpeeds(0);
  delay(100);
  setLeftBackward(); // set left motor to run backward
  setBothSpeeds(DEFAULT_SPEED); // set speeds to 1/2 default
  delay(10 * deg); // allow time for the bot to turn
                   // turn time is approx 5ms per degree
  straightForward(); // resume driving forward at default speed
}

void turnRight(int deg) {
  setBothSpeeds(0);
  delay(100);
  setRightBackward(); // set right motor to run backward
  setBothSpeeds(DEFAULT_SPEED); // set speeds to 1/2 default
  delay(10 * deg); // allow time for the bot to turn
                      // turn time is approx 5ms per degree
  straightForward(); // resume driving forward at default speed
}

Sketch Notes

First, the reaction matrix is stored in an array of bytes called cases. The bits in each byte represent a sensor. The right sensor is represented by the least significant bit, the front is the second, and the left is the third least significant bit. There is room in each byte for five more sensors, but we’re only using three.

I think defining the reaction matrix is pretty self explanatory. The only slightly tough thing is building the thisCase byte based on the sensor inputs. We start with a byte consisting of all zeroes: 0b00000000. Then, if the reading from sensor i is less than the reaction distance, we want to change bit i to 1 (we want to flip bit i). We start with 0b00000001 and shift it to the left by i places using the bitwise shift operator: <<. This creates a byte containing a 1 followed by i zeroes. For instance:

int a = 1 << 2; // a is equal to 0b00000100 (or 4)

Then we use the bitwise OR operator to flip the appropriate bit in the thisCase byte. thisCase |= (1 << i);  flips bit i in the thisCase byte.

Once the thisCase byte is built it only needs to be compared to every element in the reaction matrix.


Coming Soon

A post on my PID motor control library!

It makes using DC motors much  easier. It requires motor encoders to provide feedback to the speed and position control routines, however. So I’ll probably do posts on using encoders and PID control as well.

 

Obstacle Avoidance

 

At long last I’ve gotten around to doing a post on obstacle avoidance! Thanks to everyone for your patience.

When I started writing this post it had been months since I had last thought about obstacle avoidance. I opened my obstacle avoidance sketch for Colin intending to use it for this post unchanged but it looked terrible. It was a kludgy mess of nested if/else statements. So I took a few hours to totally rewrite my code and make it more efficient and readable. Revisiting old programs can be a great opportunity to reevaluate previous work.

Anyway, the basic idea behind obstacle avoidance is pretty simple. Colin, my mobile robot, can sense objects around himself. If he gets too close to an object he turns away and goes off in another direction. Easy, right?

Wiring Diagram
Program Design
Example Sketch
Video


Preliminaries

We’ll be using HC-SR04 ultrasonic sensors for this tutorial. If you’ve never used ultrasonic sensors before you should take a look at my tutorial. This tutorial also uses DC motors. If you’ve never used DC motors with an Arduino before, take a look at my motor control tutorial.

There is one problem to address right away: the configuration of our sensors. HC-SR04 sensors have a roughly 30° field of view, so with just one sensor Colin won’t be able to see anything to his sides.

Others have solved this problem by mounting a sensor on a servo so the sensor rotates and sweeps out a larger field of view. This instructable is a good example of the technique.

For my purposes it was easier to use an array of three sensors, however. With sensors facing forward, left and right Colin gets a pretty good picture of what’s around him. He still has blind spots at +45° and -45° though, so I’m planning on adding two more sensors.

top


Wiring Diagram

Wiring up the sensors and motors is pretty simple. We really just have to combine the wiring diagrams from the motor control tutorial and the ultrasonic sensor tutorial. Wire the components per the diagram below and you’ll be in good shape.

Wiring diagram for obstacle avoidanceIt’s come to my attention that, on some displays, the color of the TRIG and ECHO wires for the left sonar sensors is very similar to the color of the +5V wire. These wires SHOULD NOT connect, however.

top


Program Design

Before we get into actually writing the obstacle avoidance program we have to decide: how should Colin react to sensor inputs?

The simplest thing we could do is have Colin turn 90º to the left whenever one of his sensors sees an obstacle in front of him. But what if there is also an obstacle to his left? He would turn directly into the obstacle on his left while trying to avoid the one in front of him. What if there is an obstacle on Colin’s left or right but no obstacle in front? Clearly there are several possibilities here.

We need to identify the set of situations Colin might encounter that require him to react. Then we need to identify what those situations look like to Colin in terms of sensor inputs. Lastly, we need to specify an action for Colin to take in each situation.

Let’s assume Colin only needs to take action when an obstacle is within a certain distance. We can call this distance, dr for reaction distance. When the distance from one or more of Colin’s sensors to the nearest obstacle is less than dr Colin needs to take action to avoid the obstacle(s). The table below breaks down the situations and sensor inputs that require Colin to react and the action Colin could take.

Reaction Matrix

Obstacle Locations Left Distance Front Distance Right Distance Response
No Obstacles > dr > dr > dr Drive forward
Front > dr < dr > dr Turn left 90°
Front and right > dr < dr < dr Turn left 90°
Front and left < dr < dr > dr Turn right 90°
Front, left and right < dr < dr < dr Turn left 180°
Left and right < dr > dr < dr Turn left 180°
Right > dr > dr < dr Turn left 45°
Left < dr > dr > dr Turn right 45°

Notice that our reaction matrix requires Colin to turn by a number of degrees for most of his reactions. But Colin has no way of knowing how far his wheels have turned, which would be required for him to know how many degrees he’s turned. The best we can do for now is set Colin’s motors to run at different speeds for a certain time interval. Through trial and error we can find the speed differential and time interval required to get a specific amount of turning. This is an approximate solution but we can’t do any better until we add encoders to Colin’s motors.

top


Example Sketch

/*
 * Obstacle avoidance example
 * by Andrew Kramer
 * 8/3/2015
 *
 * Uses 3 ultrasonic sensors to determine
 * if nearest obstacle is too close.
 * Outlines 8 possible cases for
 * positions of obstacles.
 */

#include <NewPing.h>

#define RH_PWM 3 // PWM pin for right hand motor
#define RH_DIR1 4 // direction control for right hand motor
                  // BIN1 pin on motor controller
#define RH_DIR2 5 // direction control for right hand motor
                    // BIN2 pin on motor controller
#define LH_PWM 9 // PWM pin for left hand motor
#define LH_DIR1 7 // direction control for right hand motor
                     // AIN1 pin on motor controller
#define LH_DIR2 8 // direction control for left hand motor
                     // AIN2 pin on motor controller
#define DEFAULT_SPEED 25 // default PWM level for the motors
#define TURN_DIST 25 // distance at which the bot will turn
#define MAX_DISTANCE 200 // max range of sonar sensors
#define NUM_SONAR 3 // number of sonar sensors
#define NUM_CASES 8 // number of reaction cases

#define MS_PER_DEGREE 10 // milliseconds per degree of turning

NewPing sonar[NUM_SONAR] = { // array of sonar sensor objects
  NewPing(13, 13, MAX_DISTANCE), // left
  NewPing(12, 12, MAX_DISTANCE), // front
  NewPing(11, 11, MAX_DISTANCE) // right
};

/*  
 *  stores a bool for each sensor (left, front, and right respectively
 *  true if nearest obstacle is within TURN_DIST
 *  true if not
 */
bool sensor[3] = {false, false, false}; 

// stores all possible sensor states
bool reactions[NUM_CASES][NUM_SONAR] = { 
   {false, false, false}, // 0: no obstacles
   {false, true, false},  // 1: obstacle in front
   {false, true, true},   // 2: obstacles front and right
   {true, true, false},   // 3: obstacles front and left
   {true, true, true},    // 4: obstacles front, left, and right
   {true, false, true},   // 5: obstacles left and right
   {false, false, true},  // 6: obstacle to the right
   {true, false, false} }; // 7: obstacle to the left

void setup() {
  for (int pin = 3; pin <= 9; pin++) {
    pinMode(pin, OUTPUT); // set pins 3 through 9 to OUTPUT
  }
  Serial.begin(9600);
}

void loop() {
  updateSensor();
  switch (compareCases()) {    
    case 0: // no obstacles
      straightForward();
      break;
    case 1: // obstacle in front
      turnLeft(90);
      break;
    case 2: // obstacles front and right
      turnLeft(90);
      break;
    case 3: // obstacles front and left
      turnRight(90);
      break;
    case 4: // obstacles front, left, and right
      turnLeft(180);
      break;
    case 5: // obstacles left and right
      turnLeft(180);
      break;
    case 6: // obstacle to the right
      turnLeft(30);
      break;
    case 7: // obstacle to the left
      turnRight(30);
      break;
  }
  delay(100);
}

void updateSensor() {
  for (int i = 0; i < NUM_SONAR; i++) {
    int dist = sonar[i].ping_cm();
    // if sonar returns 0 nearest obstacle is out of range
    if (dist == 0) sensor[i] = false;
    else sensor[i] = dist < TURN_DIST;
  }
}

int compareCases() {
  for (int i = 0; i < NUM_CASES; i++) {
    bool flag = true;
    for (int j = 0; j < NUM_SONAR; j++) {
      if (reactions[i][j] != sensor[j]) flag = false;
    }
    if (flag) return i;
  }
}

void setLeftForward() {
  digitalWrite(LH_DIR1, HIGH);
  digitalWrite(LH_DIR2, LOW);
}

void setRightForward() {
  digitalWrite(RH_DIR1, HIGH);
  digitalWrite(RH_DIR2, LOW);
}

void setBothForward() {
  setLeftForward();
  setRightForward();
}

void setLeftBackward() {
  digitalWrite(LH_DIR1, LOW);
  digitalWrite(LH_DIR2, HIGH);
}



void setRightBackward() {
  digitalWrite(RH_DIR1, LOW);
  digitalWrite(RH_DIR2, HIGH);
}

void setBothBackward() {
  setRightBackward();
  setLeftBackward();
}

void setLeftSpeed(int speed) {
  analogWrite(LH_PWM, speed);
}

void setRightSpeed(int speed) {
  analogWrite(RH_PWM, speed);
}

void setBothSpeeds(int speed) {
  setLeftSpeed(speed);
  setRightSpeed(speed);
}

// sets direction of both motors to forward and sets both speeds
// to default speed
void straightForward() {
  setBothForward();
  setBothSpeeds(DEFAULT_SPEED);
}

void turnLeft(int deg) {
  setBothSpeeds(0);
  delay(100); // delay to allow motors to stop before direction change
  setLeftBackward();
  setBothSpeeds(DEFAULT_SPEED);
  delay(MS_PER_DEGREE * deg); // allow time for the bot to turn
  straightForward(); // resume driving forward at default speed
}

void turnRight(int deg) {
  setBothSpeeds(0);
  delay(100); // delay to allow motors to stop before direction change 
  setRightBackward(); 
  setBothSpeeds(DEFAULT_SPEED); 
  delay(MS_PER_DEGREE * deg); // allow time for the bot to turn
  straightForward(); // resume driving forward at default speed
}

You probably won’t be able to load up the code on your differential drive robot and run it, even if you have it wired properly. Depending on how you have the motors wired, one or both of them might run backward. To fix this you should just swap the values in the DIR1 and DIR2 fields for the problematic motor. Also, you may have to adjust the value of MS_PER_DEGREE to get accurate turning.

You’ll notice most of the code in the above sketch is devoted to simply controlling the two motors: setting motor directions, PWM levels, etc. In fact, very little of the above sketch codes for higher level behaviors like deciding when and how to react to obstacles. This makes the sketch more difficult to read and it will only get worse when we add encoders to the mix.

To fix this I’ve developed a motor control library. Encapsulating the motor control code in separate motor objects allows us to focus on programming high-level behaviors without worrying too much about how we’re controlling the motors. I’ll present my motor control library (and make it available for download) in my next post.

top


Video

Below you can see a video of Colin running the above example sketch.

You’ll notice that Colin always stops before making a turn. This slows him down and makes his behavior appear jerky. New methods could be added to the sketch to allow Colin to turn while still moving forward in some situations. These methods would simply slow down the wheel on the inside of the turn rather than reversing it. This could be useful when Colin approaches an obstacle obliquely. In this case only a minor course correction is required so a small, smooth turn is better than stopping, rotating, and starting forward again.

That’s all for today. Check back in a couple of weeks for posts on my motor control library and a refinement to obstacle avoidance that uses bit math!

top

Ultrasonic Sensors

If you’ve been through my simple motor control tutorial you now know how to control the input voltage to DC motors using an Arduino. In a differential drive robot like Colin, this means driving in a (nearly) straight line, a circle, or some other preset path. It would be more interesting if we could make a robot that reacts to its surroundings. For this we need sensors and ultrasonic sonar sensors are a good place to start.

Ultrasonic sensors determine the distance between themselves and the nearest obstacle by emitting an ultrasound pulse and timing how long it takes for that pulse to be reflected off the nearest obstacle and back to the sensor. They are easy to use, accurate, and they can be extremely cheap.

They do have limitations, however. Sound reflects best off of hard, smooth objects, so soft or uneven surfaces are not detected very well. Most kinds of fabric are basically invisible to an ultrasonic sensor. Reflection is another problem. To work properly, the sensor needs an obstacle to reflect its emitted pulse straight back to the sensor. If the sensor isn’t aimed straight at a perpendicular surface, the reflected pulse could miss the sensor entirely. In this case the sensor would not detect the obstacle. The datasheets for most sensors specify an angle within which the sensor will be able to reliably detect obstacles.

HC-SR04 sensor

HC-SR04 sensor

For this tutorial I’ll be using an HC-SR04 sensor. The datasheet for the sensor can be found here. Note that it will only reliably detect obstacles within +/- 15 degrees from perpendicular to the face of the sensor. The HC-SR04 definitely isn’t the best ultrasonic sensor out there. More accurate, longer range sensors exist. You cannot, however, find a cheaper ultrasonic sensor than the HC-SR04. If you shop around on Amazon you can find them for $1-3 apiece.

Before I get into the actual tutorial I should mention I’m making use of Tim Eckel’s NewPing library for Arduino here. It makes using these sensors a complete snap. You should definitely take a look at the tutorial on the Arduino Playground in addition to this one. If you don’t have experience installing or using new libraries in Arduino, take a look at this tutorial.


Single Sensor Example

We’ll start out by wiring up and testing a single HC-SR04 sensor. You’ll want to wire it up as in the diagram below.

Wiring diagram for a single HC-SR04 sensor.

Wiring diagram for a single HC-SR04 sensor.

Note the Trig and Echo pins on the sensor can be connected to different pins on the Arduino. The NewPing library makes it possible to connect the Trig and Echo pins to the same Arduino pin. This is great because it means the HC-SR04 doesn’t occupy as many I/O pins.

Note also there is no power source pictured on the above diagram. For the purposes of this exercise we can power the Arduino through its USB port, so no external power source is required.

Example Sketch

#include <NewPing.h>

#define TRIGGER_PIN 2
#define ECHO_PIN 2
#define MAX_DISTANCE 200 // max distance the sensor will return

NewPing sonar(TRIGGER_PIN, ECHO_PIN, MAX_DISTANCE); // declare a NewPing object

void setup() {
  Serial.begin(115200);
}

void loop() {
  delay(50);
  int uS = sonar.ping(); 
  Serial.print("Ping: ");
  Serial.print(uS / US_ROUNDTRIP_CM); // convert ping time to distance in cm
  Serial.println("cm");
}

Load the above sketch on your Arduino and open a terminal window. Make sure the baud rate of the window is the same as the baud rate in your Serial.begin();  statement.

When initializing a NewPing object use the following form: NewPing(triggerPin, echoPin, distanceLimit); If you’re using the same pin for the trigger and echo just set triggerPin and echoPin to the same pin.

The example sketch will display the distance from the sensor to the nearest obstacle in centimeters every 50 milliseconds. If the nearest obstacle is beyond the distanceLimit or if the nearest obstacle is not detectable then sonar.Ping() will return 0.

Play around with the single sensor for a bit. Have some fun with it. Eventually start to realize how limited a single sensor is. A robot that can only see directly in front of itself can’t see obstacles at its sides. So if it approaches an obstacle at an angle, the obstacle won’t be detected. So what should we do? Add more sensors, of course!


Three Sensor Example

Now that we’ve got a single sensor up and running it should be a simple matter to add two more sensors. It can be useful to have an array of sensors when we’re trying to detect obstacles around a mobile robot like Colin. Wire the sensors up as in the diagram below.

Wiring diagram for three HC-SR04 sensors.

Wiring diagram for three HC-SR04 sensors.

Example Code

#include <NewPing.h>

// trigger and echo pins for each sensor
#define SONAR1 2
#define SONAR2 3
#define SONAR3 4
#define MAX_DISTANCE 200 // maximum distance for sensors
#define NUM_SONAR 3 // number of sonar sensors

NewPing sonar[NUM_SONAR] = { // array of sonar sensor objects
  NewPing(SONAR1, SONAR1, MAX_DISTANCE),
  NewPing(SONAR2, SONAR2, MAX_DISTANCE),
  NewPing(SONAR3, SONAR3, MAX_DISTANCE)
};

int distance[NUM_SONAR]; // array stores distances for each
                         // sensor in cm

void setup() {
  Serial.begin(115200);
}

void loop() {
  delay(50);
  updateSonar(); // update the distance array
  // print all distances
  Serial.print("Sonar 1: ");
  Serial.print(distance[0]);
  Serial.print("  Sonar 2: ");
  Serial.print(distance[1]);
  Serial.print("  Sonar 3: ");
  Serial.println(distance[2]);
}

// takes a new reading from each sensor and updates the
// distance array
void updateSonar() {
  for (int i = 0; i < NUM_SONAR; i++) {
    distance[i] = sonar[i].ping_cm(); // update distance
    // sonar sensors return 0 if no obstacle is detected
    // change distance to max value if no obstacle is detected
    if (distance[i] == 0)
      distance[i] = MAX_DISTANCE;
  }
}

Note the step in the updateSonar() method that checks if the distance returned by sonar[i].ping_cm() is 0, meaning the nearest obstacle is probably out of range. Technically it’s possible there is an obstacle 0cm away from the sensor, but that is not very likely. If the value returned by sonar[i].ping_cm() is 0, we switch it to MAX_DISTANCE .

Now we can control 3 sonar sensors simultaneously. We can use this in conjunction with our ability to control motors to make a robot that has autonomous behavior! One of the simpler things we can do with these two tools is program an obstacle avoidance routine. If that interests you stay tuned for my next tutorial, because I’m planning to do in on that very topic!

First Pictures of Colin

I have an increasing number of friends and family members who use Facebook mainly as a vehicle for posting pictures of their toddlers. I try to counterbalance that by posting lots of pictures of Colin. Here’s Colin at 11 months old:

Isn't he adorable?

11 month old Colin!

Look at his cute little bluetooth radio!

Look at his cute little bluetooth radio!

I’m going to add more ultrasonic sensors soon, but this is what he looks like for now.

So far I’ve gotten him to avoid obstacles, follow walls, and I’m working on a PID motor control library. I’ll be doing a post with the technical details of motor control very soon. Stay tuned!