rpwr-assignments

This repository contains my work for the Robot Programming with ROS course during the Summer semester 2025. It includes four structured assignments and a final project focused on robot control and navigation using ROS 2 and Gazebo.

Assignments Overview

1. Git, Linux & Python Basics (01_git-linux-python) Introduces fundamental tools for robotics development:

   • Basic Git usage and version control workflows
   • Linux shell scripting and package management

2. Coordinate Frames & TF (02_coordinates-tf) Covers:

   • Homogeneous transforms and coordinate math

   • Introduction to the ROS tf2 library

   • Static and dynamic transforms for simulated robots

3. ROS Communication (03_ros) Focuses on:

   • Creating ROS 2 nodes in Python

   • Publishing/subscribing to topics

   • Services, parameters, and launch files

4. Navigation & SLAM (04_navigation) Involves:

   • Running SLAM and path planning on Turtlebot

   • Obstacle detection and local/global planning

   • Testing in simulation using nav2 stack

   • Sloved A wall Follower chalenge in the simulation

Final Project — Autonomous Turtlebot Behaviors (internship/)

For the final project, our group implemented three distinct behaviors in ROS 2 that combine perception, control, and real-time decision making in a simulated Turtlebot3:

Lane Keeping Assist

Goal: Allow user control via PS3 controller, while enabling the robot to autonomously avoid collisions with hallway walls or obstacles.

Approach:

LIDAR-based region mapping: The robot divides its surroundings into 5 angular zones: left, fleft, front, fright, and right. It computes the mean distance within each zone to assess proximity to walls or obstacles.

Reactive FSM (Finite State Machine): Based on proximity analysis, the robot switches between:

• Find the wall

• Turn left

• Turn right

• Follow the wall

Smooth navigation: It avoids abrupt stops or oscillations by applying state-driven angular corrections and reducing linear speed only when necessary.

Throttle PS3 commands: Velocity commands from the PS3 controller are throttled and blended with the robot’s autonomous corrections to allow safe control.

Knock-Knock (Door Detection & Response)

Goal: Allow the robot to detect whether a door is open or closed, and move forward only when the path is clear.

Approach:

• Robust door detection using LIDAR scan analysis across a window of ~40 degrees in front of the robot.

• Noise handling and normalization:

  • Invalid points (e.g. too close or too far) are capped or discarded.

  • Applied safety rules: ignore spurious small obstacles, smooth decision over time.

• Stable open-check mechanism: The robot only moves forward after the opening is consistently clear for several consecutive scans (required_stable_count = 4).

• Automatic environment detection: Supports both simulated and real robot setups by detecting topic names and switching accordingly.

Laser Scan Filtering

Goal: Filter noisy or irrelevant points from the /scan topic in real time to produce cleaner, more consistent LIDAR data for use by downstream behaviors such as navigation, obstacle detection, and wall following.

Approach:

Grouped filtering: LIDAR ranges are grouped in sliding windows (group_size = 3), and each group is aggregated using a configurable statistic: mean, median, min, or max. This reduces angular resolution and smooths scan data.

Blind spot masking: Angular sectors known to be unreliable (e.g., −15° to −5°, +5° to +15°) are force-cleared by setting their ranges to infinity (inf), effectively ignoring them during processing.

Outlier removal: Local spatial outliers are filtered using a sliding median window. Any point that differs from the local median by more than outlier_threshold = 0.3 is discarded as noise.

Range capping: Points below min_range = 0.15m or above max_range = 3.5m are treated as invalid and ignored, helping suppress reflective artifacts and distant clutter.

Demo & Presentation

📄 Final Report (PDF)
▶️ Watch the demo