ApexNav Real-world Test Example

This release contains trajectory-generation and MPC control code. Using the supplied scripts, you can perform a real-world test within the Habitat simulator, and the implementation is readily portable to a physical mobile robot.

Dependencies

Please follow the main README to set up the ApexNav environment and the Habitat simulator before running this example.

Quick Start

All following commands should be run in the apexnav conda environment

ROS Compilation

catkin_make -DPYTHON_EXECUTABLE=/usr/bin/python3

Run VLM servers

Run each of the following commands in a separate terminal:

python -m vlm.detector.grounding_dino --port 12181
python -m vlm.itm.blip2itm --port 12182
python -m vlm.segmentor.sam --port 12183
python -m vlm.detector.yolov7 --port 12184

Launch visualization and main algorithm

source ./devel/setup.bash && roslaunch exploration_manager rviz_traj.launch   # RViz visualization
source ./devel/setup.bash && roslaunch exploration_manager exploration_traj.launch   # ApexNav main algorithm

Run Habitat simulator (velocity-control version)

source ./devel/setup.bash && python habitat_vel_control.py

Run the real-world example node

source ./devel/setup.bash && python ./real_world_test_example/real_world_test_habitat.py

Visualization with RViz

Manually Set Goal in RViz

You can use RViz's "2D Pose Estimate" tool to set a target pose so the robot can navigate to a specified location.

Autonomous Object Navigation

You can use RViz's "2D Nav Goal" tool to start autonomous exploration.

Acknowledgment

We would like to acknowledge the contributions of the following projects:

• REMANI-Planner: Real-time Whole-body Motion Planning for Mobile Manipulators.