CS 371 - Likelihood Field Sensor Model

## Area of Focus

Our focus on this project is on improving the sensor model and parameter tuning.

## Hypothesis

• Modified the calculation way of probability from + to * for each particle will increase the aggressiveness of the model
• The initial value of probability of each particle have impacts on the speed and accuracy
• Using log function will give more smoothness to the instability of probability of particles therefore will moderate the aggressiveness

## Testing Metric

We would conduct five trials each for original and modified algorithm and keep track of the number of times they successfully localize.

## Compare of the origin and the modified algorithm

Original Modified with * Modified with * and log
Trail 1 Success Fail Success
Trail 2 Fail Fail Success
Trail 3 Fail Fail Success
Trail 4 Success Fail Success
Trail 5 Success Fail Success

## Data (Trails outcome)

From the represented test data, it shows that the original function has 3 successful localization and 2 fail. The modified with only * algorithm fail all the time as the particles allocate so fast that once they all allocate , there will be no way for the algorithm to adjust the position with the particles.

## Algorithm

In the modified algorithm, q is changed to 3 and will return the log(q) instead of q. The reason increase q to 3 is because it will enlarge the weight between different particles as it keeps multiplying pz. Therefore, increasing q will give certain particle much larger weight and when resampling particles, particles will more likely fall on those with higher weight particles and it will increase the speed of building up possible position clusters. However, if the particles allocate too fast, then there is no way to adjust the positions. The situation happens when only times pz, particles will allocate together within three to five moves. Therefore, introducing log will help to smooth and reduce the difference within different particles here. As particles allocate in a shorter time, the time modified algorithm takes will be shorter than the time original one takes.

## Possible Future Focus

• Deal with the uncertainty of the map. As the map uses in the experiment is based on the scanner, therefore, it will contains certain degree of uncertainties. Therefore, count in the uncertainty of the map can increase the accuracy of the localization with sensor model.
• Deal with the symmetric map. As sensor model uses calculated distance to estimate the position in the map. Thus, a symmetric map or map contains certain similar corners can affect the localization of the machines. It is another important thing need to focus.