EECE 5554
Robotics Sensing and Navigation

Lab 4: Dead Reckoning with
an IMU, GPS, and LiDAR

Dr. Xian Li
Assistant Teaching Professor
Department of Electrical and Computer Engineering

Today

What Lab 4 asks, and why
The car and its two computers
Before the drive: the launch file and the dry run
In the car: roles, recordings, and the car files
The analysis, including the new parts
What to submit

How long can you dead-reckon?

In a tunnel, under a bridge, or between tall buildings, GNSS can vanish for minutes. A car then has to estimate its own motion. Each sensor in Lab 4 is good at one part of that, and fails in its own way.

RTK GNSS

Absolute position, to a few centimeters, once a second.

Drops to float or standalone near buildings and trees
Gone in a tunnel

IMU

Rotation, smoothly, 40 to 100 times a second.

Heading drifts slowly
Velocity from the accelerometer drifts fast: every error is integrated twice

LiDAR

How the car moved between two scans, from the walls, trees, and poles it sees.

Fails where there is nothing to match, or the scene moves
Its own turns drift slowly

You will measure each one against RTK, and see why real cars combine all three.

The Mustang Mach-E

Sensor	Rate
Ouster OS1-64 LiDAR	10 scans/s, 65,536 points each
Ouster's built-in IMU	100 Hz
The car's own IMU	100 Hz
The car's own GPS	1 Hz, with its time
Wheel speeds, steering	50 to 100 Hz
Your RTK and VN-100	1 Hz and 40 Hz, on your laptop

The car PC runs ROS 2 Humble. The camera and radar are recorded too, but Lab 4 does not use them.

Two computers, never connected

Each computer times its own bag by its own clock. The data lead and the car PC operator start and stop together by calling out to each other, and Part 5.6 lines the two clocks up afterward.

Five parts

Part	What you do	Where
1. One launch file	`nav_bringup` starts your Lab 2 RTK driver and your Lab 3 VN-100 driver together	At home, against the emulators
2. The dry run	The whole data collection on foot: `walk_circle` and `walk_square`, until the self-check says READY TO DRIVE: YES	A park, with your team
3. The drive	Calibration circles and a 2 to 3 km loop: `car_circles` and `car_drive`, on both computers	The Mach-E, near Kerry Park
4. The tools	Reading your bags and the two car files	The notebook
5. The analysis	Heading, speed, and path from the IMU, then two clocks, LiDAR odometry, and an RTK outage	The notebook

The drive needs the car, booked by your team. Everything else you can do on your own time.

Part 1: one launch file

nav_bringup/launch/nav.launch.py

"""Start the Lab 2 RTK driver and the Lab 3 VN-100 driver together."""
import os

from ament_index_python.packages import get_package_share_directory
from launch import LaunchDescription
from launch.actions import DeclareLaunchArgument, IncludeLaunchDescription
from launch.launch_description_sources import PythonLaunchDescriptionSource
from launch.substitutions import LaunchConfiguration


def generate_launch_description():
    gps_launch = os.path.join(get_package_share_directory('gps_driver'),
                              'launch', 'rtk_driver.launch.py')
    return LaunchDescription([
        DeclareLaunchArgument('gps_port', default_value='/dev/ttyACM0'),
        IncludeLaunchDescription(
            PythonLaunchDescriptionSource(gps_launch),
            launch_arguments={'port': LaunchConfiguration('gps_port')}.items()),
        # TODO: declare imu_port, and include vn_driver's launch file
    ])

The GPS half is in the worked example. The VN-100 half is yours.
Each include passes its own port, so the two drivers can never end up on the same device.
The self-check runs both emulators, twice, with the ports swapped.

On the real sensors, use the /dev/serial/by-id/ names. They never change.

Part 2: the dry run on foot

walk_circle

Four slow circles around a landmark, about 10 m across, with the VN-100 held level. Practice for the calibration circles in the car.

walk_square

A 25 m square, with a 20 s still start. Practice for the drive, and a short, simple bag for testing your analysis.

Record with your laptop's WiFi off, as in the car: nmcli radio wifi off
Run check_bag on every bag before you leave the park. A bad bag is cheap to redo there
Commit both bags, then run the self-check. It must say READY TO DRIVE: YES before your team gets the car

The dry run is where problems are found: a loose CAT-6 loop, a wrong port, a dead battery. The car is not.

Part 3: The Drive

What happens in the car, step by step

Roles in the car

Driver

The instructor. Drives, and nothing else: never looks at a screen, and nobody asks them to.

Car PC operator

Types the car's commands on the car PC when the data lead calls for them. At the end, makes and copies the car files.

Data lead

Runs the laptop, calls the start and stop of each recording, and watches both topic rates and the fix quality.

Sensor lead

Mounts and checks the VN-100, the antenna, and the receiver, and owns the cables.

Navigator

Calls the route, keeps the hotspot on, and writes every stop and anything unusual, with the time.

With three people

The sensor lead also navigates. Pick roles before you leave, and write them in the field handout.

What to bring

Laptop, fully charged, with READY TO DRIVE: YES
A spare laptop with tested drivers
VN-100 and its USB cable
RTK kit: receiver, GNSS antenna, WiFi antenna, CAT-6 loop, USB-C cable
Phone with the hotspot from Lab 2, charged

A drive formatted exFAT for your car files: an SSD or a USB stick
Tape measure, strong tape or straps, a level app
The field handout, printed, and a pen

Print the field handout, EECE5554_Lab4_field_handout.pdf. The car has no internet, and the handout has every step, a checklist, troubleshooting, and a page for your notes.

Safety first

Everyone wears a seat belt. We follow every traffic law, and yield to pedestrians, cyclists, and the streetcar
Nobody touches a sensor, cable, or mount while the car moves. If something comes loose, we stop somewhere legal and safe first
Secure everything before the car moves: in a hard stop, a loose IMU or laptop is a projectile
Nothing on the dashboard, near an airbag, near the pedals, or in the driver's view
Anyone can stop the drive at any time, and nobody argues. A data set can be recollected

Step 10: mount, and measure

VN-100: on a flat, rigid surface near the centerline, level, x axis straight ahead, taped down, away from speakers and phones
Antenna: on the roof mount, cable over a rear door seal, never through a window, taped to the roof
Receiver: on the floor or rear seat, WiFi antenna on, CAT-6 loop in, USB plugs taped

Measure x_c, y_c, x_a, and y_a from base_link, in cm, with y positive to the right. Take a photo of each.

The terminals you need

Where	Terminal	Start it with	Used for
Laptop	1: drivers	source ~/EECE5554/Lab4/ros2_ws/install/setup.bash	Your `nav.launch.py`, running the whole trip
Laptop	2: recording	the same line, then cd ~/EECE5554/Lab4/data	Your `ros2 bag record`
Laptop	3: watching	the same line	`ros2 topic hz /imu` and `/gps`
Car PC	the car	Staff opens it	Only `~/lab4/lab4_check.sh` and `~/lab4/lab4_record.sh`

The car PC operator types only the commands in the field handout. The car's own setup must never change.

Step 11: before the car moves

Laptop WiFi off: nmcli radio wifi off. The receiver keeps its corrections through the hotspot
Drivers running with both by-id ports: /imu near 40 Hz, /gps near 1 Hz
A real fix: fix_quality 4 or 5, outside the garage, after a few minutes
Car still: /imu linear acceleration z near -9.8, because the VN-100's z axis points down
On the car PC: ~/lab4/lab4_check.sh ends with CAR READY: YES
The checklist in the field handout

car PC

~/lab4/lab4_check.sh

/front_lidar/points      10.0  OK
/front_lidar/imu        100.1  OK
/vehicle/wheel_speeds   100.0  OK
...
Free disk space: 180 GB (OK)
CAR READY: YES

One recording, two computers

The car's recording starts first and stops last, so it covers all of yours. The data lead calls each step out loud, and the car PC operator answers when it is done.

Step 12: the calibration circles

Where: a flat, open lot, near the route or not, away from parked cars, light poles, and buildings. Not a slope: the vertical field would leak into the fit
Size: 12 to 20 m across, at walking pace. Tighter is fine; the Mach-E's tightest is about 12 m
How: 20 s still, 5 full 360 degree circles, then 10 s still
Afterward: nothing in the car moves until the drive is done

car PC, then laptop terminal 2

~/lab4/lab4_record.sh start team3 car_circles
ros2 bag record -o car_circles /gps /imu

# 20 s still, 5 circles, 10 s still, then:
# laptop: Ctrl+C, and "Laptop stopped."
~/lab4/lab4_record.sh stop

These circles calibrate the magnetometer, and the same turns line up the two clocks in Part 5.6.

Step 13: the drive

The route: a 2 to 3 km loop on the flatter streets on top of Queen Anne Hill, near Kerry Park, starting and ending at the same spot
At least 10 turns, a few stops, and one hill on purpose. No tunnels, freeways, or garages
20 s still at the start, for the accelerometer bias, and 20 s still at the end
The navigator logs every stop and anything unusual, with the time

car PC, then laptop terminal 2

~/lab4/lab4_record.sh start team3 car_drive
ros2 bag record -o car_drive /gps /imu

~/lab4/lab4_record.sh status   # watch it grow

# at the end: laptop Ctrl+C, then
~/lab4/lab4_record.sh stop

A rate drops to zero? Finish the loop, fix it parked, and drive it again as car_drive_2, on both computers.

Step 14: before you hand the car back

car PC

~/lab4/lab4_record.sh list team3
~/lab4/lab4_record.sh extract team3
# plug your team's drive into the car PC, then:
~/lab4/lab4_record.sh copy team3

your laptop, later

mkdir -p ~/EECE5554/Lab4/data/car
cp /media/$USER/<drive>/EECE5554_Lab4_team3/*.npz \
   ~/EECE5554/Lab4/data/car/

ros2 bag info on both laptop bags: both topics, about 40 IMU and 1 GPS message per second
extract turns each 30 GB car bag into a 10 MB car file, in a few minutes
copy finds your drive, copies the files, and ejects it. --raw also copies the whole bags, if the drive has 40 GB free
Everything of yours out of the car, WiFi back on, problems reported

Commit the two car files, and give a copy to each teammate. Never commit the car's own bags.

If something goes wrong

Problem	What to do
A rate drops to zero	Keep driving normally, finish the loop, fix it parked, and record the loop again
Something comes loose	Do not touch it while the car moves. Stop somewhere legal and safe first
fix_quality stays at 1	Corrections are not arriving: hotspot on, CAT-6 loop in, Corrections tab live
A driver dies at launch	Device busy: close minicom. Permission denied: the dialout group
CAR READY: NO	Tell staff, and do not drive
`start` says the bag exists	It never overwrites. Use the name it suggests, such as car_drive_2, on the laptop too
`copy` cannot write to the drive	Format it exFAT. A Mac's own format, APFS, does not work on the car PC
Anyone feels unsafe	Stop the drive. A data set can be recollected

Part 5: the analysis

Section	The question	Figures
5.1 Magnetometer	Hard and soft iron, on foot and in the car	1 to 5
5.2 Heading	A complementary filter: the gyro for fast changes, the magnetometer for the long run	6, 7
5.3 Velocity	The accelerometer against RTK and the car's own speed, with bias, hill, and stop corrections	8
5.4 Motion model	Where the VN-100 sits, from the data	9
5.5 Dead reckoning	The IMU track against RTK: how long within 2 m, and within 10 m	10
5.6 Two clocks	Line up the laptop and the car PC by matching two gyros	11
5.7 LiDAR odometry	The car's path from scan matching, with an ICP you write	12
5.8 An RTK outage	IMU, LiDAR, and the two together, through 90 s without GNSS	13

Try it: line up two gyros

Part 5.6

Shift the laptop's gyro

L-6.00 s

Correlation0

True Lhidden

A made-up drive: turns of different lengths, a car gyro at 100 Hz with z up, and a laptop gyro at 40 Hz with z down, whose recording started a few seconds later.

LiDAR odometry

Part 5.7

Two scans, half a second apart, as x and y points 1 to 3 m above the ground
Move the new scan by a guess, and pair each point with its nearest neighbor in the old one
Find the turn and shift that best fit the pairs, with an SVD
Repeat until it settles: that is ICP, iterative closest point
Chain the steps into the car's path, and compare it with RTK

The guess matters: give ICP the gyro's turn and the car's speed.

On a simulated drive, LiDAR odometry drifted about 1 percent of the distance driven, mostly through its heading.

An RTK outage

Part 5.8

Pick 90 s of car_drive with a fixed RTK, pretend it is lost, and estimate the path three ways from the last good fix. The distance from RTK after 90 s, over three stretches of a simulated drive, on a log scale:

The LiDAR measures distance well and the gyro measures turns well. Together, they beat either one alone. Your own numbers go in Question 10.

What to submit

GitHub: your workspace, four laptop bags, two car files in data/car/, the analysis notebook, the README with the results table and answers, and the self-check report
Canvas: one PDF with Figures 1 to 13
Never commit build/, install/, log/, videos, or the car's own bags

Item	Points
Software and data sets, from the self-check	25
Figures 1 to 13	49
Results table	5
Questions A1, A2, and 1 to 10	41
Total	120

Lab 4 is 20 percent of the final grade. The LiDAR, the clocks, and the outage are 36 of the 120 points.

Where to find help

The instructions notebook, EECE5554_Lab4_instructions.ipynb: every step, the tools, and a troubleshooting appendix
The worked example, EECE5554_Lab4_example.md: the launch package, a passing self-check, the car files, and the README format
The field handout, EECE5554_Lab4_field_handout.pdf: the drive, step by step, printed
The self-check, ./verify_lab4.sh: run it early and often
Book the car: [[ BOOKING LINK ]]

Stuck for more than 30 minutes? Post the exact error and what you tried, or come to office hours.

Three takeaways

1Rehearse on foot, then drive once. READY TO DRIVE: YES comes before the car, and check_bag runs before you leave each site.

2Two computers, one call. The car's recording starts first and stops last, the data lead calls each step, and the clocks are lined up afterward from the data.

3No sensor wins alone. RTK is absolute, the IMU knows turns, the LiDAR knows distance. You will measure how long each lasts without the others.