Skip to content

ARDev1161/robot_orchestrator

Repository files navigation

Robot Orchestrator Workspace Bundle

This folder aggregates the core packages required to run the robot LLM planner stack on CM3588 and companion hosts. It is intentionally not a ROS package; it is a curated bundle of related packages that are built together in the same workspace.

Contents

  • robot_planner - Planning, PDDL aggregation, and PlanSys2 integration (goal gateway + action nodes).
  • ros2_rknn_llm - RKLLM-based ROS 2 node that turns user requests into structured actions.
  • robot_llm_orchestrator - Launch orchestration for planner + LLM.
  • mapannotator - Provides /pddl/map (problem template) used by pddl_aggregator.
  • rc_agent - Auxiliary robot agent utilities.
  • plansys2_upf_plugin - PlanSys2 plugin dependencies.
  • gscam2, ros-gst-bridge - Optional vision streaming components.

Build

From the workspace root:

cd ~/ros2_ws
colcon build --symlink-install
source install/setup.bash

If you only want the core stack:

colcon build --packages-select robot_planner ros2_rknn_llm robot_llm_orchestrator

Run

Launch the full stack (planner + PlanSys2 + LLM):

ros2 launch robot_llm_orchestrator robot_llm_orchestrator.launch.py \
  model_path:=$HOME/qwen3-vl-4b-instruct_w8a8_rk3588.rkllm

Key topics:

  • /rkllm_node/user_text - User input text
  • /rkllm_node/assistant_text - Structured JSON replies
  • /pddl/problem - Aggregated problem PDDL
  • /robot_planner/goal - Goals pushed to PlanSys2
  • /robot_planner/goal_status - Goal accept/reject status

Troubleshooting

  • If /pddl/problem is empty, ensure mapannotator publishes /pddl/map.
  • If PlanSys2 fails with PDDL parse errors, check the domain requirements (e.g., :numeric-fluents).
  • If RKLLM fails to allocate NPU memory, ensure no stale rkllm_node processes are running.

Notes

  • This folder is a convenience bundle, not a standalone ROS package.
  • Use consistent ROS_DOMAIN_ID and RMW_IMPLEMENTATION across machines for DDS discovery.

Node Interaction Graph

flow

About

Robot controlling via LLM

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

 
 
 

Contributors