Forward collision warning system
A warning threshold built from real driver braking data, then running live on a LiDAR robot.
Outcome
Working warning demonstrated live on a robot
Key outcomes
- 01Processed radar data from 66 braking runs
- 02Time-to-collision threshold of 1.03 s
- 03Live demo on a µlab robot
The problem
A collision warning has to come early enough for the driver to react, but not so early that it annoys them into ignoring it. The threshold should come from how real drivers brake.
What I did
I was part of the whole process. We processed radar and vehicle kinematics data from 66 braking runs in Python: synchronising time, rejecting invalid runs and filling gaps in the radar range using the vehicle's kinematics. From the 15th percentile of driver braking behaviour we set a time-to-collision warning threshold of 1.03 s.
We then implemented the warning with LiDAR on a µlab robot, compensating for the sensor's offset from the robot's front.
The result
The warning ran on the robot and was demonstrated live to the examiner.
Next project
Powertrain modelling of ICE and electric vehicles
Hiring?