Quickstart

This section assumes you have built the workspace and sourced it. See Installation for full setup details.

1. Make sure the LLM/VLM endpoints are reachable

If you want LLM/VLM behavior, ensure the endpoints configured in your profiles are running. If you do not have an LLM/VLM available, set agent_type:=baseline-0 or agent_type:=baseline-1 when launching.

2. Run an example

Rock Breaking

ros2 launch kea_rock_breaking kea_rock_breaking_planner.launch.py

Moleworks

ros2 launch kea_moleworks kea_moleworks_planner.launch.py \
  action_params_file:=`ros2 pkg prefix kea_moleworks`/share/kea_moleworks/params/action_parameters_success.yaml

3. Choose an agent type

The agent_type launch argument controls how KEA reacts to failures:

• baseline-0: no recovery logic (reports failure).
• baseline-1: automatically replan on failure.
• baseline-llm: use the LLM plan solver plugin for planning.
• kea-agent: use the LLM plan oversight agent for recovery decisions.

Example:

ros2 launch kea_rock_breaking kea_rock_breaking_planner.launch.py agent_type:=baseline-1

4. Start problems on demand

If you launch with auto_start:=false, the controller waits until it receives a problem request. You can send one via a service call:

ros2 service call /start_problem_execution kea_msgs/srv/StartProblemExecution "{problem: '/path/to/problem.pddl'}"

See the Configuration section for additional launch arguments and problem input formats.