Skip to main content

Pluto Robot

1. Overview

The software runs on the Raspberry Pi 5 and is built on ROS2 (Robot Operating System), a framework in which small programs (“nodes”) exchange data through named channels (“topics”). Each sensor and function is one node, so parts can be replaced or tested separately.

Item Purpose Status
Ubuntu 24.04 Server Operating system Working
ROS2 Jazzy Middleware Working
Chassis driver (ugv_driver) Drive commands, odometry, battery data Working
Manual control (keyboard) Teleoperation Working
Robot model and bringup Sensor positions, one-command startup Working
LiDAR driver 2D scan data Working
BNO085 driver External heading sensor Working
GPS driver Position data Working
EKF sensor fusion Combines wheel and heading data Working, heading problem open
SLAM (mapping) Map building Not resolved
Nav2 (autonomous navigation) Path planning and driving Not started (needs a usable map)
Camera and pedestrian detection Obstacle and pedestrian detection Deferred
GPS waypoint following Fallback for autonomous driving Not started (only backup)

First start of the Raspberry with not-finally used OS

2. Core Infrastructure & Drivers

Operating system: Ubuntu 24.04 LTS Server (64-bit)

Why: The Waveshare reference software targets Raspberry Pi OS and ROS2 Humble. We chose Ubuntu 24.04 because it is the platform that the ROS2 version Jazzy is built for and it is more lightweight than raspberry os.

First steps: Installation on a microSD card, headless (no screen), operated over SSH.

Raspberry with installed micro SD card

Deployment: The Pi is reached over SSH, through a phone hotspot in the field or via Ethernet. Work runs inside tmux (see Quick start guide), so processes survive a dropped connection.

Problems:

    • The Pi’s serial port was reserved for Bluetooth by default. Bluetooth was disabled. 

    • The serial device was missing on one kernel version.  Resolved.

    • The camera software (libcamera) does not work on Ubuntu 24.04 yet. Open, see the camera item. Backup the SD card before approaching an installation!

 

Middleware: ROS2 Jazzy

Why: ROS2 provides ready-made, tested components for the main tasks (sensor drivers, sensor fusion, mapping, navigation), so these did not have to be written from scratch. Jazzy matches Ubuntu 24.04.

First steps: Installation of ROS2 with the Nav2, SLAM and sensor fusion packages.

Deployment: Our own packages live in one workspace (robot_ws). All nodes are started through a launch file or manually in separate tmux windows.

Problems:

    • The Waveshare ROS2 workspace is for Humble and could not be used directly. It served as reference only.

    • Edits to configuration or robot-model files only take effect after a rebuild (colcon build) and a restart. Resolved. 

    • Two nodes publishing the same topic or transform corrupt the data silently. Resolved.

 

Chassis driver (ugv_driver, own development)

Why: The Waveshare driver code does not fit our OS and ROS version, so we wrote our own node. It converts ROS drive commands to the chassis’s serial format and returns wheel speeds and battery voltage.

First steps: Based on the Waveshare reference code. At startup the node configures the chassis and sends a stop command. It also continuously resends drive commands, because the chassis stops by itself after 3s without a command.

Deployment: Publishes wheel odometry, the built-in IMU data and the battery voltage. Wheel odometry distance was checked against a measured distance (6.33 m reported vs. about 6.3 m measured) on our testing track in IPEK hallway. A watchdog stops the chassis if power was lost and restored.

Calibration track in IPEK

Problems:

    • Duplicate publication of the odometry transform corrupted the map. Now only the EKF publishes it. Resolved.

    • Chassis power cut-out under high payload. Resolved with a buffer capacitor.

    • Wheel slip. Resolved with glued on tires to the rims.

 

Manual control (keyboard teleoperation)

Why: A safe, simple way to drive the platform during development and mapping, and a manual override.

First steps: An existing keyboard tool was too slow for the safety timeout. We wrote our own WASD node.

Deployment: Runs in its own tmux window over SSH. Speeds during testing were limited to 0.2 m/s.

Problems: Reconnecting over SSH injected stray characters into the controls. Resolved. A wireless gamepad would be better but was not purchased.

Manual control panel on waveshare phone app – but you can also manually control via Ubuntu

Robot model and bringup

Why: The software needs to know where each sensor sits on the robot. This is defined once in a robot model file (URDF). A single startup file launches LiDAR, sensor fusion and mapping together.

Deployment: Sensor positions are fixed in the model. The LiDAR’s scanning height is 365 mm.

Problems:

    • The LiDAR manufacturer’s launch file published a wrong transform. We always start through our own bringup file. 

    • The LiDAR mount was rotated 90°, which the model did not reflect. Corrected. 

    • The model must be re-measured and the software restarted whenever a sensor mount is moved. 

 

3. Sensor Integration & Evaluation

Driver Why Notes and problems Status
LiDAR (ldlidar_stl_ros2) Official manufacturer driver The Waveshare fallback code was not used. Running
BNO085 IMU (from source) No ready-made Jazzy package Three patches were needed (bus address, topic name, covariance values). They are lost if the driver is re-downloaded. Running
GPS (nmea_navsat_driver) Standard ROS2 package Position confirmed outdoors. Update rate is low (about 1 Hz) and velocity output was not verified. Running

Left: Lidar on top of pre-build, Right: new external IMU and GPS

4. High-Level Navigation & Open Issues

Sensor fusion (EKF, robot_localization)

Why: Wheel odometry is accurate in distance but not in heading (skid-steer). The EKF (extended Kalman filter) combines wheel data for distance with the BNO085 for heading.

First steps & Deployment: Runs at 30 Hz (higher frequence was tested as well without improvement) and publishes the filtered position and the transform from odometry to robot.

Problems: A configuration error let wheel-based heading influence the result in spite of the BNO085. Fixed the configurations and maps improved. The consistent curvature of outdoor maps remains unexplained (suspected magnetic disturbance, not confirmed).

Status: Not resolved. We did not pursue it further because of limited time.

Map resulting from mapping run on a straight course

Mapping (SLAM, slam_toolbox)

Why: A map of the surroundings is the basis for Nav2.

First steps: Indoor and outdoor test drives with teleoperation.

Deployment: Needs a manual activation step after every start.

Problems: Maps were fragmented. Fixing the EKF error made single map parts consistent, but they do not merge into one map, and outdoor maps curve. Several parameters were changed at once, so no cause could be isolated.

Status: Not resolved. We did not pursue it further because of limited time.

Mapping result in the hallway of the IPEK on our testing track

 

Autonomous navigation (Nav2)

Why: Standard ROS2 package for path planning and obstacle avoidance.

Status: Not started, because it requires a usable continuous map.

 

GPS waypoint following (fallback)

Why: Chosen as the fallback for autonomous driving, because mapping did not work and time was limited.

Status: Not implemented

 

Camera and pedestrian detection

Why: Planned to detect pedestrians (YOLOv8 nano model).

Status: Deferred. The camera software does not work on Ubuntu 24.04 yet.

Camera modul is installed but not in use – make a backup before installing the driver