go_fetch_ws

CLA Course Project

This repository is a complete ROS 2 workspace for the "Go Fetch!" demo on NRG's B3V0 robot platform.

For this demo, Spot will navigate between predefined waypoints, autonomously detect and grasp tennis balls, and deposit them into a designated AprilTag-marked collection bin.

The full integrated demo combines autonomous navigation, ball detection, approach planning, and manipulation into a single behavior tree workflow.

Repository Structure

Custom repositories inside src/:

• spot_tennis_demo → Launch files, YOLO detection, and perception pipeline
• tennis_demo_behaviorized → Behavior tree logic, navigation decisions, and manipulation sequencing

Requirements

• Intended to run on B3V0 shared robot environment
• User must have access to the Docker daemon on the B3V0 machine
• AprilTag-tagged bin must be visible to Spot's left after undocking (can be changed by editing Waypoint A)
• Waypoints B & C must be obstacle-free (set near the Glovebox)

Clone

git clone --recursive git@github.com:andonbreitenfeld/go_fetch_ws.git
cd go_fetch_ws

If you forgot to clone recursively:

git submodule update --init --recursive

Build

cd go_fetch_ws
colcon build
source install/setup.bash

Environment Variable

Before running the demo, ensure the following variable is added to your bashrc:

echo 'export SPOT_ACCESSORIES="ARM RL_KIT"' >> ~/.bashrc
source ~/.bashrc

Launch Sequence

Open 6 separate terminals and run the following commands in order:

Terminal 1: Spot Drivers & Core Systems

ros2 launch spot_tennis_demo demo_bringup.launch.py

Terminal 2: Navigation Stack

ros2 launch spot_navigation bringup_launch.py

Terminal 3: Manipulation Stack

ros2 launch spot_moveit_config move_group.launch.py

Terminal 4: YOLO Detection (Docker)

cd src/spot_tennis_demo/docker
docker compose up yolo

Terminal 5: Perception Pipeline

ros2 launch spot_tennis_demo perception.launch.py

Terminal 6: Behavior Tree

ros2 run tennis_demo_behaviorized run_tennis_tree

What Happens

1. Startup: Spot claims, powers on, undocks, and navigates to startup position
2. Bin Localization: Continuously attempts to locate the AprilTag bin until successful
3. Patrol & Monitor: Cycles between waypoints while monitoring for stable ball detections
4. Fetch Sequence: Upon detecting a ball:
   • Computes approach goal and navigates to it
   • Executes grasp sequence
   • Returns to bin and deposits ball
5. Resume Patrol: Returns to waypoint cycling until next ball is detected

System Architecture

• Perception: YOLO 3D detections → Custom BallSelector filter node → Stable Ball Pose
• Planning: Custom ComputeApproachGoal behaviors for ball & bin
• Navigation: Used premade WalkToPose behavior for precise positioning
• Manipulation: Custom PickBall and DropBall behaviors

License

MIT

Authors

• Andon Breitenfeld
• Clara Summerford
• Luke Pronga