This National Robotics Initiative project will promote the progress of science, advance the national prosperity, welfare, and security; by developing novel buoyancy-assisted collaborative robots that are cheap, safe, and never fall down. The current state of robotics systems is not ready to be deployable to human daily environments, which are challenging for a robot to navigate, especially for legged robots. While they are naturally well-suited for human environments by design, legged robots are often too expensive to scale up for a multi-agent setup, heavy and dangerous to operate near humans, and hard to guarantee stable locomotion. To this end, the ultimate goal of this research is to build collaborative robots that are cheap, safe, and never fall down by exploiting buoyancy to defy a significant amount of gravity This research has the potential for significant impact in enabling deployed locomotion for safe robotic systems interactively assisting humans in daily environments. It provides a principled way for leveraging large amounts of safe and scalable hardware designs with recent advances in machine learning techniques to develop compact representations that are transferable across different robotic systems and human environments. This will innovate how one can utilize a deployable multi-agent system in disaster relief zones and large outdoor environments. The project team will invite general public participation by publicizing the hardware designs and open-sourcing all the deployed software infrastructure. The grant will also support a competition for middle and high school students using the developed low-cost platforms with the goal to foster students? interest in science, technology, engineering and math (STEM).
Creating a new class of locomotion systems has two major challenges: designing a new hardware that is cheap and safe and developing an algorithm for locomotion and collaboration. In order to address these two challenges, this grant will support development of a novel framework that (1) addresses a fundamentally new family of legged robots, namely buoyancy assisted robots (BARs), which are constantly stood-up and highly light-weight by leveraging the inherent "lifting power" of buoyancy; and (2) empowers BARs with reliable locomotion and collaboration skills using deep reinforcement learning algorithms. To this end, the research team will perform interrelated research thrusts centered around the goal of safe, scalable multi-robot systems: developing multiple BARs with each unique locomotion style, learning primitive motor skills for a single agent, extending individually trained agents to orchestrate in a large-scale multi-agent friendly environment.
This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.
