clarkli86/carnd-unscented-kalman-filter-project

Self-Driving Car Nanodegree Program Starter Code for the Unscented Kalman Filter Project

Unscented Kalman Filter Project Starter Code

Self-Driving Car Engineer Nanodegree Program

In this project utilize an Unscented Kalman Filter to estimate the state of a moving object of interest with noisy lidar and radar measurements. Passing the project requires obtaining RMSE values that are lower that the tolerance outlined in the project rubric.

How to build and run the project

For how to build and run the project, please refer to the readme at https://github.com/udacity/CarND-Unscented-Kalman-Filter-Project/blob/master/README.md

Final Result

The final result is better than the result obtained with EKF (https://github.com/clarkli86/CarND-Extended-Kalman-Filter-Project). RMSE metrics especially vx and vy are much lower. This is because Unscented Kalman Filter tends to work better with non-linear transformations.

Parameter Tuning and NIS Analysis

std_a and std_yawdd had to be turned to meet the requirements in the rubic (https://review.udacity.com/#!/projects/284/view). The default values (30 and 30) seem to keep the consistency in UKF (Unscented Kalman Filter) but does not acquire the required performance.

After looking at the statistics of simulation data, I decided to apply noises that are close to the standard deviation of longitudinal acceleration and yaw acceleration.

The final chosen parameters at listed in the following table:

Parameter	Value	Note
std_a	0.285	stdev * 4
std_yawdd	0.391	stdev * 4

When these parameters were small, it was observed that RMSE values were out of spec, and the population of NIS samples greater than threshold is more than 5%. The NIS output suggests that the filter has underestimated the uncertainty, thus the noise covariances need to be increased.

The final chosen parameters demonstrats a resonable NIS plot.