The software for the Domino Robot project is the largest solo software project I have done to date. It runs across three computers and contains almost 15,000 lines of code.

The architecture of the project itself evolved rather organically as the scope and complexity of the project grew. There are many design choices and libraries that are present because development started on an Arduino and was later ported to a Raspberry Pi when more compute was needed. There are also many simplifying assumptions made to make the code simpler to write. However, since this code needs to run reliably for hours on end, there are also many assumptions that add complexity to try and ensure the software is robust to errors and can ‘fail safely’.

Below is a diagram that shows all of the major pieces of the software.

Domino Robot Software Architecture (click to enlarge).

Master

The master program is responsible for all of the high level planning of the domino field and sends step-by-step instructions to the robot in order to place the dominos correctly in the field. It runs on a remote laptop that connects to the robot over wifi.

Field Planner

The FieldPlanner takes in an input image of what the final field should look like. It then generates the domino field by scaling the image to the appropriate size (1 pixel = 1 domino), matching colors in the image with the available domino colors we have, and splitting the image up into 15x20 'tiles' that represent the group of dominos the robot can put down at once. These tiles are then sequenced such that the robot never has to drive near the tiles for very long and the next tile location the robot has to get to is always accessible. Finally, for each tile a sequence of 'actions' is generated which are used as the primary means of communicating with and controlling the robot. These actions are things like Load Dominos, Move to X Position, Move Using Vision, Place Dominos, Request Status, etc. These action sequences form cycles where each cycle places one tile into the field. All the cycles are essentially identical other than the positions the robot moves to. This group of cycles can be saved to and loaded from disk as a 'plan'.

GUI and Runtime Manager

Once there is a plan, the main master program runs and is responsible for handling a few things: monitoring and communicating with the robot, monitoring and executing the plan, and providing a GUI for the user to interface with the robot and the plan. The GUI is a relatively simple UI that has buttons for sending commands and modifying the plan, status outputs for the robot and plan, and a visualization of the robot position in the world. The RuntimeManager handles the plan execution by iterating through cycles and actions, sending each action to the robot, and then waiting for confirmation that the action is completed. The RuntimeManager can hold a collection of interfaces to various robots (which are mostly just a few layers of wrappers around a TCP socket) as the original plan was to have the master software interface with multiple robots and possibly multiple base stations, but in the end it was scaled back to just a single robot.

Robot Client

The master-robot communication protocol works sort of like a client-server protocol, but slightly worse. The robot acts as a server where it waits for actions to be sent to it. It will immediately acknowledge the action request and then proceed to carry out the specified action. The master sends regular status requests to the robot, which will respond with the current status, including the state of the current action which the master uses to track the robot’s progress. This allows for the communication handling to be a bit simpler as the master always sends one message and receives exactly one message back immediately after. There is no need to handle asynchronous communication which greatly simplifies the code.

Robot

The robot software is the most complex part of the system and it runs on the Raspberry Pi onboard the robot. Most objects on the robot operate in a cooperative multitasking mode where none of them block execution for any significant amount of time. Methods that do block or take a long time are run in separate threads. The top level robot object implements a simple loop that checks for a new command, starts a new command if needed, updates the currently running command, and checks if the current command is completed.

Robot Server

The RobotServer is responsible for all the communication with the Master software. Internally, it uses a multi-threaded wrapper around a TCP socket to avoid blocking while waiting for client activity. The server parses incoming packets as JSON and then checks for a type field which must be present in all commands to indicate the type of the command. Certain commands also contain additional data from Master such as a position or status. The Robot Server then returns the command type and any additional data to the main robot loop so that the correct actions can be taken for the given command.

Status Updater

The StatusUpdater maintains the current status of the robot and many other classes take this object by reference so that they can all update the global status. When the Master software requests the robot status, this module serializes the status to JSON and returns it back to the Master.

Tray Controller

The TrayController provides an interface for controlling the tray to load and place dominos. It maintains small state machines for each of the tray actions (PLACE, LOAD, and TRAY_INIT) and sequences individual subcommands (such as moving the lifter to a specific spot, waiting for some amount of time, or opening or closing the latch that holds the dominos) to send to the Motor Driver over a USB Serial connection.

Marvelmind Wrapper

The MarvelmindWrapper handles interfacing with the Marvelmind Library and gets the current position from the Marvelmind sensors. It then provides this information to the top level robot object which routes the data to the RobotController module.

Camera Tracker

The CameraTracker is responsible for tracking the position of the robot computed from the markers on the ground. Within the CameraTracker, two CameraPipeline objects run image processing for each of the cameras. Each CameraPipeline spawns a thread because it takes about 100 ms to fully process one image. This pipeline does some initial filtering and thresholding and then uses blob detection to find the marker in the image. The marker is then converted from image coordinates to physical coordinates by using the intrinsic and extrinsic properties of the camera. When the marker pose is returned from each camera for the current image, the CameraTracker computes an estimate of the robot pose relative to the marker positions.

Robot Controller

The RobotController handles all of the motion control and driving of the robot. It maintains global localization using odometry and position information from the Marvelmind sensors. It also can instantiate different control modes for different types of movement. Each control mode generates a command velocity at each time step and implements its own check for trajectory completion. The command velocity at each time step is sent to the motor controller over a USB Serial connection. There is a lot more information about the control system in the article I wrote about it.

Global Localization

Global localization is handled by a Kalman Filter that fuses local odometry from the motors and global position estimates from the Marvelmind sensors. The Kalman Filter itself is very simple as it only maintains the 3x3 position estimate but there is a little bit of extra work that goes into processing the position update step from the Marvelmind. Since the Marvelmind readings can lag when the robot is moving, the estimated variance of the measurement is scaled with the velocity of the robot. This is not an accurate model as the variance is not truly Gaussian, but it works well enough.

Position Control Mode

The Position Control Mode handles most of the driving modes for the robot. It generates a point to point trajectory (more info on how the trajectory generation works) and then uses simple PID controllers along with feedback from the localization system to generate velocity commands for each axis that are then sent to the motor controller.

Vision Control Mode

The Vision Control Mode is used for all motions where the markers on the floor need to be tracked. It does not use the global localization system and instead only considers the relative displacement of the markers from their target. In practice, this is still handled by computing an 'estimated pose' of the robot so that the trajectory generation system can work in the same fashion, but in essence it is just driving the markers in the image to a target location. This control mode also handles its own 'localization' using a second Kalman Filter that combines the wheel odometry with the 'position estimates' provided by the CameraTracker. Similarly to the Position Control Mode, it uses simple PID controllers to generate velocity commands for each axis to send to the motor controller.

Motor Driver

The Motor Driver runs on a Teknic ClearCore which was generously provided by Teknic for the project (along with the drive motors for the base). This board functions very much like an 'Industrial Arduino' so much so that Teknic actually provides an Arduino-compatible library so it can be programmed directly from the Arduino IDE. This board is responsible for all of the real time driving of the motors (drive motors and the lifter motor), the servo that drops the dominos, the optical endstop for the lifter, and the buttons to operate the lifter manually.

Lifter Action Handler

Communication with the Raspberry Pi that is specific to the lifter gets routed to the lifter action handler which has simple state machines for operating the servo, moving the lifter to a specific height, and homing the lifter.

Robot Velocity Handler

Velocity commands for the base are routed to the robot velocity handler which takes in a cartesian velocity in the local robot frame, converts it to wheel speeds, and then updates the commanded velocity for each motor. It also returns an odometry estimate from the motor, but due to how the ClearPath motors work, this isn't actually a true encoder reading from the motors. However, the motors themselves are extremely accurate and follow the velocity commands extremely well so not reading directly from an encoder is not really a huge problem.

That is a high level overview of how the software for the Domino Robot works. If you are interested in looking at the actual code, you can find it all on Github.

46 Comments