Active Goal Recognition on the Insect Hotel Scenario

This repository contains the source code to replicate the demo scenario described in:

J.C. Saborío, M. Vinci, O. Lima, S. Stock, L. Niecksch, M. Günther, A. Sung, J. Hertzberg and M. Atzmüller. "Uncertainty-Resilient Active Intention Recognition for Robotic Assistants". In Proceedings of ECMR 2025 (to appear).

Video

We are preparing a detailed video showcasing the Mobipick robot in the Insect Hotel domain. A link will be added here as soon as it is ready.

Installation

ROS Noetic and Python 3 are assumed to be available on your system. vcs and wstool will be installed when needed:

./install-deps.sh
./build.sh

Setup and Usage

Real Robot Demo

Start up the robot according to the (DFKI internal) instructions on the wiki, then on the robot launch:

roslaunch mobipick_bringup mobipick_bringup_both.launch  # already part of the startup instructions
roslaunch pbr_dope dope.launch
roslaunch insect_hotel_bringup bringup.launch   # world_config:=insect_hotel   (default)

On the Jetson Orin launch the camera driver and object detection:

roslaunch yolo6d_cpp yolo6d_with_cameras.launch

To start the demo, launch on the robot:

roslaunch insect_hotel_bringup intention_recognition.launch

Robot Simulation

Launches the robot in Gazebo with simulated worker cameras.

roslaunch insect_hotel_bringup demo_sim.launch   # world_config:=insect_hotel_tables_spread   (default)
roslaunch insect_hotel_bringup intention_recognition.launch simulation:=true

Worker Simulation

The behavior of the worker is simulated by moving insect hotel parts from the storage area to the assembly area. Inside the insect_hotel_bringup/launch/includes/intention_recognition.launch these lines modify the simulated workers behavior. The following parameters can be changed:

• hotel_type: Specify hotel type 1 or 2, type 1 uses red side panel and dark green back panel, type 2 uses orange side panel and black back panel
• worker_time_between_actions: Time in seconds between taking parts from the storage
• random_worker_actions: If true takes parts from storage randomly, if false takes parts in fixed assembly order
• prob_to_skip_action: Probability for the worker to do nothing and skip an assembly action
• prob_to_take_wrong_part: Probability for the worker to take a wrong part, e.g. a part that does not belong to the insect hotel type that is currently being built
• prob_to_fix_wrong_part: Probability to fix the workers mistakes, if any occured

Plan Visualization

Install and source the dot_graph_visualization rqt plugin, then call it with:

rqt --standalone dot_graph_visualization

Advanced Documentation

World Configs

This repository defines two so-called world_configs (i.e., arrangements of tables, objects etc.):

1. insect_hotel: The regular arrangement of tables in our physical robot lab, with the insect hotel parts spread over the tables. Available in reality and simulation. This is the default for the real robot demo.
2. insect_hotel_tables_spread: The same as above, but the tables are spread apart further. Does not reflect the actual table arrangement in the lab, so this world_config is only available in simulation. This is the default for simulation.

The reason for introducing the second world_config is the following: In reality, we use the internal move_base instance that is running on the internal PC of the MiR platform. This move_base instance can deal with the regular, crowded table arrangement. Obviously, we cannot directly use the MiR PC in simulation, so we had to replicate the move_base config in the mir_navigation package for use in simulation. Unfortunately, this config does not include some proprietary parts of the MiR configuration, so it sometimes struggles in crowded environments. To work around this issue, we spread the tables further apart for simulation only.

Switching between the two is done by passing the argument world_config:=insect_hotel... to bringup.launch (for running on the real robot) or demo_sim.launch (for simulation).

TL;DR: In reality, use world_config:=insect_hotel. In simulation, use world_config:=insect_hotel_tables_spread. These are already the defaults.