Home

Localizability-aware Exploration Planning

Uncertainty-aware Receding Horizon Exploration Planner

The Uncertainty-aware RHEM planner performs planning for autonomous exploration, while ensuring that viewpoint-to-viewpoint trajectories minimize localization & mapping uncertainty. Its operation depends on closing the loop between two pipelines:

Planning pipeline:

Implemented in the bsp_planner ("belief-space propagation" planner) package, it is responsible for maintaining and updating the probabilistic volumetric map of the environment, and for performing a 2-layer planning optimization:

• The first planning layer gives the next-best-viewpoint that leads towards an increased mapping knowledge (volumetric exploration & mapping probability improvement).
• The second level gives the best-trajectory to reach that selected viewpoint (in terms of localization & mapping uncertainty).

Estimation & Propagation pipeline:

The rhem_planner stack provides a localization and mapping pipeline, augmented to perform both tasks. The included pipeline is based on a modified version of the Robust Visual-Inertial Odometry pipeline. It is responsible for:

• Calculating and maintaining the real-time localization & mapping state and statistics (Estimation).
• Forward-simulation of the process given an initial state and statistics and a sample trajectory (Propagation).

Loop-Closure scheme:

The Planning pipeline performs RRT-based sampling to reach the 1st layer's next-best-viewpoint. Each RRT vertex carries the Estimation & Propagation pipeline state (current pose, landmarks, and associated statistics). The communication between the two pipelines is performed with ROS service-based data transactions:

• The Planning pipeline requests and acquires from the Estimation pipeline the current state, which is used to populate the root vertex "belief".
• For each RRT vertex, the Planning pipeline transfers its parent's vertex "belief" to the Propagation pipeline. During the first request, also the current 3D occupancy map is tranferred (used by the Propagation pipeline to evaluate feature visibility).
• The Propagation pipeline performs forward-simulation of the localization and mapping process, and responds with the resulting "belief" state at the target vertex. The Planning pipeline stores this and re-uses it in subsequent propagation calls.
• The process is repeated, until the 1-st layer's next-best-viewpoint is reached via a trajectory that gives better localization & mapping statistics than the "straight-path" motion towards it.

Launch the RHEM Planner

To launch the RHEM pipeline directly you can use the following launch file:

roslaunch rhem_planner rhem_exploration.launch

This will bring up:

• bsp_planner: The planning pipeline node.
• rovio_node: The estimation pipeline node.
• rovio_bsp_node: The propagation pipeline node.

To trigger a planning iteration a call to the respective bsp_planner service is required. This can be directly performed from the terminal:

rosservice call /bsp_planner '{header: {stamp: now, frame_id: world}}'

A demo case for use with experimental data is also provided. Launch files and configuration settings are under the rhem_demo folder, and you have to download a bag file with visual-inertial data and put it under the same folder. You can then launch the RHEM pipeline demo with:

roslaunch rhem_demo/rhem_demo_icra17.launch

More detailed explanations about the demo can be found under the respective "How-to use Demo" section of this wiki. Also, more details about the experimental setup, regarding the exploration scenario and the employed aerial robot and onboard sensors can be found at the relative publication:

@inproceedings{papachristos2017uncertainty,
  title={Uncertainty-aware Receding Horizon Exploration and Mapping Using Aerial Robots},
  author={Papachristos, Christos and Khattak, Shehryar and Alexis, Kostas},
  booktitle={2017 IEEE International Conference on Robotics and Automation (ICRA)},
  pages={to appear},
  year={2017},
  organization={IEEE}
}