This project implements global localization using a particle filter (Monte Carlo Localization) for an indoor robot operating with odometry and a laser rangefinder on a pre-built 2D map. The system runs at 40Hz on an 80x80m map.
Each particle's pose is updated using the odometry motion model as defined in Chapter 5 of [1]. The proposal distribution is derived from the robot's odometry and control input: [ p(x_t^m | u_t, x_{t-1}^m) ]
The laser beam measurement model is implemented as described in Section 6.3 of [1]. The model, ( p(z_t | x_t, m) ), is a mixture of four probability densities, each accounting for a specific type of measurement error:
- Hit Model (p_hit) Represents accurate measurements that align with the expected range based on the map.
- Short Model (p_short): Captures unexpected short-range readings caused by obstacles closer than expected.
- Max Model (p_max): Accounts for max-range readings when no obstacle is detected within the laser's range.
- Random Model (p_rand): Models random noise in the measurements.
The overall probability ( p(z_t | x_t, m) ) is a weighted sum of these components, ensuring robust handling of various measurement scenarios.
Particles for the next time step are sampled based on their weights from the current time step using a low-variance resampling technique. This minimizes the variance in particle filtering and improves the estimator of the robot's true belief.
-
(GPU)Preprocessed Ray Map: GPU-accelerated ray-casting on the unoccupied map space, performed once during initialization for efficiency. Remark: It takes less than 30sec to generate the map (80x80m^2)
-
Adaptive Particle Count: Dynamically adjusts the number of particles based on the variance of particle weights. Enable with
--adaptive.
-
Robot Kidnapping Detection: Detects robot kidnapping events using the mean-to-max weight ratio. Enable with
--detect_kidnap.
Navigate to the code directory and execute the following command:
cd code/
python3 main.py --visualize- Enable Visualization :
--visualize - Set Number of Particles :
--num_particles=<value> - Path to Log File :
--path_to_log=<path> - Enable Adaptive Sampling :
--adaptive - Enable Robot Kidnap Detection :
--detect_kidnap
[1]: Probabilistic Robotics, by Sebastian Thrun, Wolfram Burgard, and Dieter Fox.
This repository contains my solution to 16833-Localization and Mapping Homework1. It is shared for educational purposes and should not be used for plagiarism or violating academic integrity policies. Please adhere to your institution's honor code.