Researchers Develop Multi-Robot Deployment Research Platform

Researchers from the University of Cambridge have developed a robotics platform designed to ease multi-robot research deployments.

The Cambridge RoboMaster platform offers optimized hardware and software solutions that can be deployed in small robotic solutions.

The system is specifically designed for the DJI RoboMaster S1, a compact four-wheeled robot that researchers can use to test new applications and concepts.

The Cambridge team claimed that out-of-the-box, the RoboMaster S1 lacks sufficient computing power and customizability for multi-robot experiments.

To enhance the S1 for multi-robot research, they upgraded the onboard computer to an Nvidia Jetson Orin NX, which increased its AI computing power. Despite changing the onboard computer, they retained the robot's compact size and agility to ensure suitability for indoor deployments.

Credit: University of Cambridge

They also integrated additional hardware to help the robots better navigate their environment, including extra sensors and a Raspberry Pi camera with a fisheye lens to improve its field of view.

On the software side, their platform includes simulation frameworks like a Vectorized Multi-Agent Simulator (VMAS) to better enable multi-unit operations. VMAS is designed to speed up the time it takes to train a group of robots to work in a multi-unit environment.

Related:Sony Develops Microsurgery Robot

The researcher's platform enabled the S1 robots to better perform multi-robot actions. They demonstrated the robots in action, with the wheeled units able to detect potential collisions and navigate optimal pathways between two direct points with improved power efficiency.

The RoboMaster platform can be accessed through GitHub, with other robotics researchers able to use its software features in their own S1 tests.

“We significantly extend the capabilities of the base platform by providing multiple options for more flexible and powerful compute solutions, onboard sensors, model-based control and sim-to-real capabilities for distributed reinforcement learning policies,” the researchers wrote in a paper.

The RoboMaster platform is an expansion of the researcher's previous work, where they employed Graph Neural Networks (GNNs) to teach multi-robot systems path planning by sharing communication through a localized network.