Progress Report 2013-02-01

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 2

Human-Robot Interaction in Unstructured Environments Progress Report February 1, 2013 Kevin DeMarco

Papers Read
Paper 1 Paper 2 Programming Finished abstract Kalman Filter in C++ Finished two Particle Filter variants. Straightforward basic Particle Filter and Monte Carlo Particle Filter. The weighting of the particles was determined by their euclidean distance from the centroid of the cluster. Unfortunately, this resulted in all the particles converging to a single particle, that did not have the same velocity as the centroid of the cluster. Since, the basic particle lter has no means for generating particles with new velocities, the nal particle to survive was merely the closest to the centroid of the cluster. This resulted in the particle eventually drifting away from the clusters centroid. To x this problem, I implemented the Monte Carlo Particle Filter (MCPF). After a particle is sampled based on its weight, two more particles are generated, but their states are randomly generated. During the next sampling iteration, only half of the particles are sampled, which results in maintaining the same number of particles after each iteration. While, this tracks the centroid extremely well, it doesnt smooth the trajectory, which is what is needed for behavior prediction. Kalman Filter Tests / Examples

Figure 1: Kalman Filter of noisy sensor measurement. Ground truth is static at 1.

To Do
ToDo 1 ToDo 2

Figure 2: Kalman Filter of noisy sensor measurement. Ground truth is a ramp.

You might also like