Platforms for Validation of Multi-Agent Autonomy
Abstract
We have developed a versatile, easy-to-use simulation engine that can be leveraged by the swarm research community to rapidly develop, evaluate, and adjust swarm algorithms for a variety of common benchmark scenarios. Our simulator is built on Microsoft AirSim and Unreal Engine, which provide support for vehicle and UAS models, together with a photorealistic graphics engine. We created an interface to allow swarm researchers to easily deploy and test their algorithms in complex simulated scenarios (such as search-and-rescue, ISR, and pursuit-evasion). This interface will allow researchers to choose the makeup of their swarm, and upload algorithms for different swarm tasks (such as patrolling, formation control, rendezvous, coverage, task allocation, etc.). The platform will maintain a library of benchmark scenarios, so that different algorithms can be easily and fairly evaluated against each other. We developed the simulator specifically to be deployed on the cloud, allowing ease of access to a wide class of researchers, obviating the need for expensive hardware and setup time, and allowing scaling to large swarms and environments. We took initial steps to enable the ability to test reinforcement learning algorithms for various swarm tasks such as formation control. We have also developed a multi-agent hardware testbed consisting of both unmanned ground vehicles (UGV) and unmanned aerial vehicles (UAV) to combine their respective advantages (such as the large payload capacity of UGVs, and the maneuverability and speed of UAVs). The testbed leverages both low-cost quadrotors (AR Drones, Crazyflies, Mambo) and advanced autonomous vehicles (Jackal UGV) available at the PIs lab, which can work collaboratively through local communications and coordination. The testbed supports modularized sensors to enable more functionalities. Users can customize different sensors for a variety of tasks by simply plugging them into the expansion port of the deck.
Document Details
- Document Type
- Technical Report
- Publication Date
- Oct 17, 2022
- Accession Number
- AD1212765
Entities
People
- Shaoshuai Mou
- VinÃcius L. de Lima
Organizations
- Purdue University