Advancing Field Exploration Using ROS-powered Robotic Vehicles

Click the above image for Project Video Demonstration

ARMS Lab, Systems and Control engineering department IIT Bombay & e-Yantra

Table of Contents

1. Introduction
2. Project Goal
3. Tech Stack
4. Setup
   • Prerequisites
   • Remote Desktop Setup
   • Interfacing Hardware
     • Raspberry Pi 5
     • BMX 160 IMU
     • Ydlidar Tmini-Pro
5. Key Challenges
6. Hardware Development
7. Control System Design
8. ROS2-Teleoperation
9. Localization and Odometry
10. Motion Planning
11. SLAM
12. Future Developement

Introduction

Field exploration robots have gained significant attention in recent years due to their potential applications in various domains such as agriculture, environmental monitoring, and autonomous driving research. These robots are designed to operate autonomously in diverse environments, collecting data and performing tasks that would otherwise be labor-intensive or hazardous for humans. This project focuses on developing a ROS-powered robotic vehicle designed for field exploration. The vehicle is capable of traversing varied ground surfaces autonomously and can be used for tasks such as autonomous field mapping, agricultural plant monitoring, and testing autonomous driving algorithms.

Project Goal

The primary goal of this project is to develop a ROS2-enabled four-wheel-drive vehicle with Ackermann steering capable of waypoint navigation and obstacle avoidance. This vehicle will have various sensors and an onboard computer to perform Simultaneous Localization and Mapping (SLAM). The project encompasses hardware development, sensor integration, firmware and algorithm development, and extensive testing and validation to ensure the vehicle’s performance and reliability in real-world scenarios.

Tech Stack

• ROS2 jazzy: Middleware for robot software development.
• Raspberry Pi 5: Onboard computer for processing and communication.
• Esp32: Microcontroller for low-level motor and servo control interfaced using ArduinoIDE
• Nav2: Navigation stack for ROS 2.
• RViz2: Visualization tool for ROS 2.

Setup

Prerequisites

• Operating System: Ubuntu 24 LTS noble or later
• ROS 2 Distribution: Jazzy Jelisco
• Hardware: Raspberry Pi5 , BMX 160 IMU , Ydlidar Tmini-Pro , Esp32 , 7V Lipo Battery
• Software: Arduino IDE, Python 3.12

Remote Desktop Setup

1. Install ROS2 Jazzy:

   After installing run this command once

   source /opt/ros/humble/setup.bash
   echo "source /opt/ros/humble/setup.bash" >> ~/.bashrc
   source ~/.bashrc
   

2. Create a workspace and clone the repository

   mkdir -p ~/ros2_ws/src      # Create a workspace , builld and source it 
   cd ~/ros2_ws
   colcon build --symlink install
   source install/local_setup.bash
   
   cd /pi_ws/src      # To clone the repo inside src folder 
   git clone https://github.com/TahsinOP/eYSIP-24_Field_Exploration_ROS_Vehicles.git
   

   To avoid sourcing everytime you open a new terminal instance run this once ,

   echo "source ~/ros2_ws/install/setup.bash" >> ~/.bashrc
   source ~/.bashrc      # source bashrc as we have made changes

Interfacing Hardware

Raspberry Pi 5

• Flash SD Card

1. Download Ubuntu 24: Download the Ubuntu 24 image for Raspberry Pi from the official Ubuntu website.
2. Flash the SD Card: Use tools like Balena Etcher/Rpi Imager to flash the Ubuntu image to a 32GB or 64GB SD card.

• Install ROS 2 Jazzy

1. Follow the ROS 2 Installation Guide: Official ROS 2 Jazzy installation instructions can be found here.

• Workspace Setup

1. Create ROS 2 Workspace:

   mkdir -p ~/pi_ws/src
   cd ~/pi_ws
   colcon build --symlink install
   source install/local_setup.bash
   
   echo "source ~/pi_ws/install/setup.bash" >> ~/.bashrc
   source ~/.bashrc      # source bashrc as we have made changes

2. Clone the repository

   cd /pi_ws/src
   git clone https://github.com/TahsinOP/eYSIP-24_Field_Exploration_ROS_Vehicles.git

• USB serial Permissions Add dialoutto groups

   sudo usermod -aG dialout $USER

BMX 160 IMU

Dependencies

1. Install smbus :

   sudo apt update
   sudo apt install python3-smbus

IMU node

2. Launch IMU node:

   ros2 launch imu_pkg imu_rviz_robot.launch.py

Logging Data

3. Check IMU data:

   ros2 topic echo /imu_raw

Ydlidar Tmini-Pro

Connect and Launch the Lidar driver

1. Ensure the YDLidar is connected to a USB port and powered on

2. Launch the YDLidar node with the following command:

   ros2 launch ydlidar_ros2_driver ydlidar_launch.py

3. Visualize Lidar scan data:

   ros2 launch rcar_viz display.launch.py

   This will launch Rviz2 with a robot model and fixed frame as "odom"

4. Optional: Run IMU Pose Estimation: To see the Odom data, you can run the following node (optional):

   ros2 run imu_pose_estimation estimator

   Alternatively, set the fixed frame to "base_link" in RViz to see laser data on the /scantopic.

Key Challenges

• Low-level Control: The primary challenge we faced in low-level control was interfacing a custom servo with five pins for which no pin diagrams or references were available online. Achieving accurate control at this level required extensive experimentation and fine-tuning.
• Localization and Odometry: We relied solely on an IMU to estimate the robot’s pose for localization and odometry. The IMU’s orientation measurements exhibited significant drift over time. Additionally, the robot’s suspension system continuously affected the accelerometer's bias.
• Motion Planning: In motion planning, traditional control algorithms used for differential or omnidirectional drives were ineffective. This required the development of an approach tailored to the dynamics of an Ackermann steering system.
• Environment Mapping: For environment mapping, we had to adapt mapping techniques to work with Ackermann kinematics. This included ensuring accurate data collection from LIDAR and integrating these readings into a coherent map. Figure 2: Vicon Odometry vs IMU Odometry

Hardware Development

• The motor actuation and steering system features a five-pin servo control using PID for precision and DC motor control for throttle.
• A custom PCB board was developed to ensure optimum power supply to the ESP32, motor drivers, and aspberry Pi.
• The mechanical design involved creating laser-cut CAD models using acrylic for a two-layered system with mounts for all hardware components including a LIDAR.
• BMX-160 IMU was interfaced and calibrated to publish accurate orientation and acceleration data on a ROS-2 network.
• Ydlidar Tmini-Pro interfaced with Raspberry Pi5 on ROS2 Jazzy to publish laser data on /scan topic

Figure 1: Prototype Vehicle

Control System Design

Figure 2: Control System Design

1. Low-Level Control

• ESP32:

  • Role: Acts as the intermediary between the motor drivers and the Raspberry Pi 5.
  • Communication: Receives angle and PWMvalues via UART Serial from the Raspberry Pi 5.
  • Outputs: Sends control signals to the motor drivers.

• Motor Drivers:

  • Role: Control the steering and throttle motors based on signals received from the ESP32.

• Servo Motor (Steering):

  • Role: Adjusts the steering angle of the vehicle.
  • Control: Receives signals from the motor driver.

• DC Motor (Throttle):

  • Role: Controls the throttle of the vehicle.
  • Control: Receives signals from the motor driver.

2. Onboard Raspberry Pi 5

• Operating System: Ubuntu 24LTS

• Optimized Pure Pursuit Algorithm:

  • Role: Calculates the desired angle for steering based on the current position and the goal.
  • Output: Angle and PWM values sent to the ESP32 for motor control.

• BMX160 IMU:

  • Role: Provides IMU data for odometry.
  • Connection: Connected via GPIO pins on Rasbperry Pi 5.
  • Data Published: /imu_data topic.

• YDLidar Tmini Pro:

  • Role: Provides LiDAR data for SLAM.
  • Connection: Connected via USB.
  • Data Published: /scan topic.

• Odometry (Modified Kalman Filter):

  • Role: Processes IMU and other Vicon car velocity (corresponding to PWM) data to provide an accurate estimate of the vehicle’s position and velocity.
  • Output: Publishes odometry data to the /odom topic.

3. Remote PC

• Operating System: ROS2 Jazzy

• SLAM (Simultaneous Localization and Mapping):

  • Components: EKF (Extended Kalman Filter) and AMCL (Adaptive Monte Carlo Localization).
  • Role: Utilizes LiDAR and odometry data to build and update a map of the environment.
  • Inputs: Receives /scan and /odom topics.

• Navigation Stack (Nav2):

  • Role: Handles path planningand navigation.
  • Inputs: Receives odometry data and processed map information.
  • Outputs: Publishes path data to the /plan topic.

• Goal Node:

  • Role: Defines the navigation goals for the vehicle.
  • Output: Publishes waypoints to the /goal_topic.

Summary

• The Low-Level Control segment, managed by the ESP32, translates commands from the Raspberry Pi into physical movements of the vehicle.
• The Onboard Raspberry Pi 5 is the central processing unit for sensor data collection and preliminary processing (IMU, LiDAR), and it runs the Pure Pursuit algorithm for immediate steering adjustments.
• The Remote PC is responsible for advanced processing tasks, including SLAM for mapping and localization, and the Nav2 stack for path planning and navigation, ensuring the vehicle reaches its defined goals efficiently.

ROS2-Teleoperation

Achieving complete low-level control of the car through keyboard commands using ROS2 for communication over a shared network is a significant milestone. This setup allows for real-time teleoperation, enabling users to control the car remotely using simple keyboard inputs. By implementing teleoperation nodes in ROS 2, commands can be sent to the robot, providing a responsive and reliable control system.

Implementation Steps:

1. Install necessary ROS 2 packages for teleoperation.
2. Write a ROS 2 node to capture keyboard inputs and publish velocity commands.
3. Configure the robot to subscribe to these velocity commands and actuate the motors accordingly.
4. Test the teleoperation setup to ensure reliable and responsive control.

How to Run the Teleop Node:

1. SSH into Raspberry Pi:

   ssh arms@192.168.0.171    # Change the IP address and Pi name  accordingly 

2. Run Subscriber Node On Pi/Remote PC

   ros2 run teleop_bot sub

3. Run Publisher Node On Pi

   ros2 run teleop_bot pub

   If u have less computation on Pi run the publisher on the remote PC using the same command stated above

Now u can control the car using WASD keys for movement :

• W- Move Forward

• A- Turn Left

• S - Move Backward

• D - Turn Right

• H - Halt

• Note: The throttle value has been kept constant for the time being, you can change the value in the subscriber script in the teleop_bot package. Also, try running the car at 25-50 PWM range for safe teleoperation

Localization and Odometry

• Modified Kalman Filter: Fused car velocity data (calculated from Vicon) with IMU accelerometer readings to get a more accurate state estimate (position and velocity) by reducing drift. The Kalman filter uses Vicon data as a reference and corrects the IMU readings accordingly giving accurate position data. For orientation, gyroscope data was used to calculate roll, pitch, and yaw after bias removal.

To run the odometry Node

1. SSH into Raspberry Pi:

   ssh arms@192.168.0.171    # Change the IP address and Pi name  accordingly 

2. Run Odom Node On remote PC

   ros2 run localization imu_odometry
   

This will give you the (x,y,yaw) data of the vehicle , and also publish data on /odom topic

Motion Planning

• Optimized Pure Pursuit Algorithm: Implemented an optimized version of this path-tracking algorithm to ensure appropriate steering angles using the lookahead distance concept. The waypoints are provided to the vehicle by the Goal_Node, which is fed into the pure-pursuit algorithm. The algorithm used Odometry data from the /Odom topic, calculates the linear and angular error and proceeds towards the waypoints effectively ( Note: The tuning parameter is lookahead_distance) Waypoint navigation

1. SSH into Raspberry Pi:

   ssh arms@192.168.0.171    # Change the IP address and Pi name  accordingly 

2. Run Goal Node On Pi

   ros2 run navigation GoalNode

3. Run Pure Pursuit Node on Pi

   ros2 run navigation pure_pursuit

   Hit enter on the goal node terminal after the pure-pursuit controller is started

SLAM

• SLAM: Mapped the environment using LiDAR scan data and fused it with odometry data. This, along with the Adaptive Monte Carlo Localization (AMCL) algorithm, provided an accurate pose (position and orientation) estimate of the car within the map. Slam_toolbox is used for mapping the environment with proper loop closure, and the Nav2 stack is used to generate local and global cost maps, generate planned paths which are finally fed in the form of waypoints to the motion planning algorithm.

1. Map the environment using slam_toolbox ( Already done and saved in the maps folder)
2. Launch the Odometery node using the steps given in the previous sections
3. Launch the Navigation and Localization launch files
4. Run the GoalNode and pure pursuit script
5. Give a 2D goal pose in Nav2 Rviz

Future Development

1. Upgrading Hardware for Outdoor Use: Enhance the hardware to function robustly in outdoor environments and incorporate GPS for improved outdoor localization.

2. Developing a Mission Planner Stack: Create a mission planner stack capable of generating and updating paths for the vehicle based on environmental variables, enabling more dynamic and adaptive route planning.

3. Implementing State-of-art Algorithms: Integrate deep learning and reinforcement learning-based algorithms to achieve more adaptive and robust navigation, enabling the vehicle to better handle diverse and unpredictable conditions. Upgrade the algorithms to better handle dynamic environments, both indoor and outdoor, as they are currently designed for static settings.

4. Enhanced Sensor Integration: Incorporate additional sensors such as RGB-depth cameras, and ultrasonic sensors to improve environmental perception, and implement advanced algorithms such as ORB-SLAM.