Reactive Wall Following Using Spline-Fitted Lidar Scans with Pure Pursuit Control
Christian (16), Ray (6-3, 18), Leo (6-4), Alicia (2)
Team 4
Developed a wall follower and safety controller on racecar hardware
Wall following is navigation via surface tracking
Controller
Sensor Data
Motors
Drive Command
Safety controller = emergency braking system
[1] Video from ANCAP Safety Ratings, full reference at end
We created a wall follower based off a pure pursuit model that maintains a desired distance from a wall with a built in safety controller
Wall Follower
01
Team 4
Linear Regression with PD Control
Pure Pursuit With Spline Tracking
Pure Pursuit With Spline Tracking
Pure Pursuit With Spline Tracking
*Implemented with AI assistance
Safety Controller
02
Team 4
Lidar Preprocessing
Object Detection
Coast is clear… keep going!
Obstacle! Stop driving!
Safety Controller Successful Stop
Testing
03
Team 4
We implemented a standard testing model that focused on testing the safety controller and the most common driving scenarios for the wall follower changing: velocity, initial angle, corners
Safety Controller Testing
Velocity (m/s) | Safety Range (m) | Measured Distance (m) | Percent Error (%) |
1 | 1.000 | 0.835 | 16.5 % |
2 | 1.000 | 0.395 | 60.5 % |
3 | 1.000 | FAILED | N/A |
3 | 2.000 | 0.970 | 51.5% |
Wall Follower Tests
RIGHT
STRAIGHT
1
LEFT
STRAIGHT
2
RIGHT
ANGLED
3
LEFT
ANGLED
4
Our wall follower had an average error of -0.0056m
Our wall follower works with extreme starting angles
Our wall follower adapts to turns
Our wall follower adapts to turns
The wall follower error was NOT dependent on velocity
Slope: -0.0063
Uncertainty (-0.0456, 0.0330)
Future Testing
02
01
Complicated paths
Dynamic Safety Controller
Lessons Learned
02
01
Naming Conventions
Record more than just laser scan
Lessons Learned
04
03
Importance of team communication
Testing Plan
Q & A