Campus conversation robot — ROS 2 workspace

A C++-first ROS 2 robot skeleton for a Raspberry Pi 5 running Ubuntu 24.04 arm64 and ROS 2 Jazzy. The source tree lives in ros2_ws/src. This is not a working autonomous robot until real sensors, encoders, motor drivers, safety hardware, and an inference runtime are integrated and tested on the robot. No sample code should be taken as a safety-certified controller.

Architecture

Package	Responsibility
motor_controller	C++ differential H-bridge driver; starts disarmed, demands fresh /scan and /odom, watchdog and latched ROS E-stop
robot_bringup	Opt-in launch of Nav2/SLAM Toolbox, camera and plugins; actuator arming is separate
memory_service	C++ persistent per-person, consent-gated memories
gpt_bridge	C++ local / Anthropic / OpenAI text chat bridge from /audio/heard to /robot/say
student_vision	Camera identity adapter for explicitly enrolled people only
feature_plugins	Allowlisted Python ROS feature plugins
terminal_input	Text-only development conversation input; not voice recognition

I interpret “SLAMR LiDAR” as LiDAR-based simultaneous localization and mapping because no LiDAR make/model was specified. SLAM Toolbox consumes a driver-provided sensor_msgs/LaserScan on /scan; Nav2 supplies planning. Neither the LiDAR driver nor wheel odometry, camera driver, speech recognition, and speech synthesis can be selected correctly without your hardware choices. No motor command should ever come from LLM output.

Install on robot

Install Ubuntu 24.04 arm64, ROS 2 Jazzy, liblgpio-dev, python3-colcon-common-extensions, ros-jazzy-navigation2, ros-jazzy-nav2-bringup, ros-jazzy-slam-toolbox, libcurl4-openssl-dev, and nlohmann-json3-dev. Install remaining ROS dependencies via rosdep. Package-specific Python dependencies are documented in their package files. Build on the Pi:

source /opt/ros/jazzy/setup.bash
cd ros2_ws
rosdep install --from-paths src --ignore-src -r -y
colcon build --symlink-install
source install/setup.bash
ros2 launch robot_bringup robot.launch.py

armed and navigation default to false; even with armed:=true, the motor node should stay stopped without fresh scan and odometry. After integrating real drivers, odometry plus odom->base_link TF, LiDAR plus base_link->laser TF, encoder feedback, a tested physical emergency-stop and correctly wired H-bridge, inspect scans in RViz and test with wheels raised. Only then opt in:

ros2 launch robot_bringup robot.launch.py navigation:=true armed:=true

Motor GPIO pin defaults are 18/23 (left forward/reverse) and 24/25 (right forward/reverse). Verify pin numbering and never attach motors directly to Pi GPIO; use a suitable H-bridge, separate supply, common ground, fuses, and a hardwired E-stop. Override pins, wheel separation and wheel speed from ROS parameters for actual hardware. The PWM motor driver is not a closed-loop velocity controller; it cannot deliver accurate Nav2 odometry by itself. The ROS /emergency_stop topic is supplementary and cannot replace a hardwired safety circuit.

Conversation and privacy

For text smoke tests run ros2 run terminal_input terminal_input_node in a second terminal and read ros2 topic echo /robot/say. To talk aloud, supply microphone STT publishing transcripts on /audio/heard and TTS consuming /robot/say; choose/measure on-device components for your latency budget. By default, the LLM endpoint comes from a separately installed local inference service; no model weights are shipped here. Alternatively select provider:=anthropic or provider:=openai with explicit allow_cloud_api:=true and the matching environment key (ANTHROPIC_API_KEY or OPENAI_API_KEY). Saved-memory export requires the separate allow_cloud_memory:=true opt-in in addition to person/session consent. See gpt_bridge setup for hidden key prompts and launch commands. Treat student identity and face embeddings as sensitive personal data: get explicit informed consent for enrollment and remembering, document retention/access/deletion, provide a no-recording mode, and obtain campus approval before deployment, especially for minors. Do not log transcripts, API keys or biometric images. Review local law and institutional policy.

ros2_ws/README.md contains component interfaces and additional setup details. Development on this Windows machine can run Python/static tests; ROS 2 and real GPIO hardware are absent, so a Pi build, integration test and safety validation remain required.