What is the difference between two-wheel and three-wheel odometry?

Two-wheel and three-wheel odometry designs differ in their sensor configuration, accuracy, cost, and how they measure heading .

Two-Wheel Odometry

Sensor Configuration: A two-wheel design uses one parallel and one perpendicular odometry pod to measure x and y movement, respectively .
Heading Measurement: Change in heading is measured using a gyroscope .
Accuracy: It offers pretty good accuracy and is cheaper than the three-wheel design, but is subject to more drift . The BNO055 IMU, often used for the gyroscope, utilizes I2C, which is slower than other I/O on the REV Hub and cannot be bulk read, leading to minute drift issues that can compound over time .

Three-Wheel Odometry

Sensor Configuration: A three-wheel design uses two parallel and one perpendicular odometry pod to measure x and y movement .
Heading Measurement: This design measures heading by calculating the difference between the two parallel wheels, forgoing the need for a gyroscope . The formula for calculating the angle (φ) is (Δxl - Δxr) / L, where Δxl and Δxr are the displacements of the left and right sensors, and L is the lateral distance between them (trackwidth) .
Accuracy: It provides relatively accurate tracking, with great accuracy in a 30-second autonomous mode . This design is often more accurate in the context of the FTC control system compared to the two-wheel design .
Cost: It is quite pricey, and tuning the heading is very important .

Both designs use "dead wheels," which are small, non-driven wheels attached to an encoder sensor, often sprung to the ground for accurate tracking . A set of three dead wheels can cost a minimum of $100 for encoders alone, before hardware .

Was this right?

Sources6

Dead Wheels > Two-Wheel Odometry Pods

Game Manual 0·CC BY-NC 4.0

Pros:

Cheaper than 3-wheel design
Pretty good accuracy
No tuning of the heading necessary
Cons:

Subject to more drift than the 3-wheel design

Pros:

Relatively accurate tracking. Great accuracy in a 30-second autonomous mode
Cons:

Quite pricey
Tuning of the heading is very important

Open Game Manual 0

Dead Wheels

Game Manual 0·CC BY-NC 4.0

Many teams in the community have converged on a unique solution that isn't seen very much outside of FTC: the use of "dead wheels," "tracking wheels," or "odometry pods" (these terms are all synonymous). These refer to small "dead" or non-driven (not powered by a motor) wheels attached to an encoder sensor. Two or three dead wheel pods are often sprung to the ground to ensure accurate tracking. The two-wheel design utilizes one parallel and one perpendicular pod (parallel and perpendicular with respect to the drive wheel axis), measuring x and y movement respectively. Change in heading is measured via a gyroscope. The three-wheel design utilizes two parallel and one perpendicular pod, measuring x and y movement respectively. However, this design forgoes the gyroscope and instead measures heading via the difference with the two parallel wheels. This is often more accurate in the context of the FTC control system because the BNO055 IMU (used for the gyroscope in the two-wheel design) utilizes I2C which is slower than the rest of the I/O on the REV Hub and cannot be bulk read. These two issues lead to minute drift issues which can compound over time, thus leading to a more inaccurate localization system when using the two-wheel design.

However, designing consistently accurate dead wheels proves to be a difficult design challenge. It is often quite pricey. A set of three dead wheels will cost a minimum of $100 for the encoders alone, prior to any hardware.

Let's go through the advantages and disadvantages of each system.

Open Game Manual 0

Odometry > Finding the Change in Position > Angle and Displacement

Game Manual 0·CC BY-NC 4.0

The displacement (or change in position) of the left sensor is Delta xl and the displacement of the right sensor is Delta xr. The lateral distance between these two sensors is called the trackwidth, notated as L. This is very important for determining angle for turning approximations. This value will need to be tuned, which means tested repeatedly and then brought to some converging value that is close to the actual measurement.

(Figure: 17508 Rising Tau's 2019/20 Skystone Bot (https://learnroadrunner.com/dead-wheels.html#three-wheel-odometry))

Deriving the value of varphi then becomes simple:

varphi = (Δ xl - Δ xr)/(L)

To perform later calculations, we need to know the displacement of the robot in the x-direction relative to its center rather than the two parallel sensors. To do this, we take the average to derive Delta xc, or the center displacement:

Δ xc = (Δ xl + Δ xr)/(2)

The final displacement we need before we can determine the change in pose is the horizontal displacement Delta xₚerp. This is the displacement of the perpendicular sensor Delta xh with a correction for forward offset F. In order to get accurate approximations, the forward offset needs to be considered. When the sensor is closer to the back, the offset is negative, but when it is closer to the front, it is positive. This is to account for the change in its position based on point-turns.

As a result of this, we can define our horizontal displacement as:

Δ x_perp = Δ xh - (F * varphi)

Note: If you do not have perpendicular sensors, which are not required if the robot cannot move in the lateral direction, Delta xₚerp is not necessary.

For this value, use 0 if you do not have a horizontal sensor.

Open Game Manual 0

Dead Wheels

Game Manual 0·CC BY-NC 4.0

Warning: This is a very niche aspect of design in FTC®. Generally it is something done by more experienced teams who have had time to repeatedly test their designs and mechanisms with software during the off-season.

The term dead wheels, tracking wheels, odometry pods, and odometry are often conflated in the FTC community. However, there are a few key differences one must keep in mind. Odometry is an umbrella term and refers to the general use of motion sensors for localization purposes. Meanwhile, dead wheels, tracking wheels, and odometry pods are all synonymous terms.

Term: Dead Wheel — A small unpowered wheel (usually an omni wheel) that tracks the distance the robot has traveled through the encoder attached to the wheel's axle.

Usually, there will be two or three wheels - one or two on the x and y axis each to track the front-back and left-right position relative to the starting point. Generally, odometry wheels are sprung so that the wheel is in contact with the floor tiles at all times to ensure accuracy.

Odometry refers to the use of motion sensors for localization. Localization is a means for being able to locate the position of the bot at some point in time. Localization is crucial in path following and advanced autonomous modes as one needs to know where they are to generate the necessary movements needed to reach a desired destination. Localization software plays a major role in odometry; however, in order to produce accurate results, reliable and accurate hardware design is a necessity.

The simplest form of odometry is drive encoder localization. This is the use of encoders measuring the rotation of motors that power the drive train. One is able to read the encoder data and feed it through the kinematic equation for that specific drive train to derive the body's velocity. Drive encoder localization is generally quite simple and easy to setup as almost all of the FTC legal motors have built-in encoders. Getting drive encoder localization setup is simply a matter of plugging in wires, no additional hardware needed.

Open Game Manual 0

Odometry

Game Manual 0·CC BY-NC 4.0

Odometry is a form of localization that uses data from sensors like encoders to derive an estimated position relative to a starting point. Localization is a means for being able to locate the position of the bot at some point in time. Odometry is especially useful in autonomous programs because it allows for easier implementation of different tasks on the field due to understanding one's position.

Open Game Manual 0

Odometry > Finding the Change in Position

Game Manual 0·CC BY-NC 4.0

In order to determine the current location of the robot and update its pose, the change must be calculated using data read from the sensors. For a robot, there will be three possible sensors that you can use: two that are parallel with the robot's body in the x-direction and one that is aligned with the y-direction of movement (perpendicular to the drive wheels).

Open Game Manual 0