Coordinated Exploration

Exploration Mapping of Unknown Environment using Coordinated Robots :

1. Single Agent Exploration

Complex underground settings present significant challenges for military and civilian first respondents. The hazards vary drastically across domains that can degrade or change over time and are often too high-risk for personnel to enter.

Objective : To Explore an unstructured and unknown environment using a Single Exploration Agent. 

Literature Review : 
[Compilation of Related Resources]
1. Frontier Based Exploration
2. Receding Horizon – Next Best View Planner
3. 2 Stage Optimised Receding Horizon Next Best View Planner

Comprehensive Resources : 
Robotic Mapping and Exploration – Cyrill Stanchniss
Efficient Algorithms and Systems for Multi-Agent Mapping and World-Scale Real-Time Localisation – Simon Lynen

The open source package of Receding Horizon – NBVP is released by ASL, ETH Zurich on GitHub. [Here].
The package is integrated with RotorS [Rotor Simulator]. Here is the installing instruction and solutions to general Issues.
P.S. : The package works well with Gazebo7 but it does not support Gazebo8 or 9 as of October, 2019.

For Hardware Implementation the package is needed to be integrated with PX4 based FCU. 
For Simulation close to real implementation : 
1. Integrate NBVP Package with PX4/avoidance package. The package PX4/avoidance is integrated with Gazebo with PX4 Firmware based UAV. 
For starters I have routed that the goal and trajectory published by NBVP to PX4/Avoidance package.
2. NBVP is highly dependent on ground truth localisation as published in Simulation. But, practically UAV localisation is not that accurate, so I integrated VINS-Fusion package for Visual Odometry.

Testing Next-Best-View-Planner with RotorS

2. UGV UAV Tracking for Failure Recovery

A 2nd Agent is introduced [UGV] for tracking UAV, which is helpful for purpose of Failure Recovery.
UGV does it’s independent exploration while staying close to UAV such that if UAV looses it’s localisation, which is fairly common in case of UAV because of loss of features, UGV tracks the UAV and helps it to re-localise. 

Maps from both the agent are shared to an external computer, where they are processed and merged. 

The UGV being used is equipped with 2D LiDAR [hokuyo 2D], and the 2D Map is Merged with the 2D transform of the 3D map from UAV, to the localisation and pose of the agents are updated accordingly.

UGV Tracking UAV

Using VINS for Visual Odometry in Exploration

3. Time Bounded Exploration with Agent Recovery

In practical Exploration, the time of operation of each agent is fixed and it is essential for them to return to their home position before their battery dies, thus there is a need for time based feasibility check at each node. 
In present approaches, the node is selected to maximise information gain, with that we add another constraint to check feasibility of return to home time from that node.
Recovery path is traced back from the node with maximum information and if the time to execute the same is within budget it is selected and discarded otherwise.

Work in Progress.

4. Multi-Agent Collaborated Exploration

Sub Parts Involved :
1. Multi Agent Path Planning : [Multi Agent RRT* Paper]
2. Decentralised SWARM operation. [multimaster_fkie]
3. Tracking and Pose Estimation for each other and Localisation & Mapping Update.
4. Optimised Recovery Plan to Return.

Two Types of Multi-Agent Exploration : 
4.1 : One-Charge Exploration 
The condition here is that there is no provision to charge the battery of agents in between exploration mission.
4.2 : Exploration with mobile Charging Station
Here UGVs agents carries extra power that can be used to charge UAVs. It becomes an Operational Research linear Optimisation problem of given X Charging station and Y UAV exploration agents what is the maximum time of exploration that can be achieved, or given T time of exploration what is the value of X and Y. [Assuming some ratio of Charging : Discharging Time. ]

Hardware Prototypes

Agricultural Automation
Hex-Rotor UAV
SWARM Quad-UAVs
SWARM - Quad-Rotors
Husky and Jackal
Husky and Jackal UGVs
Autonomous Dune Buggy
Dunne Buggy (Diesel)

Leave a Reply

Your email address will not be published. Required fields are marked *