New tool guides planning and control of robots working in uncertain environments

By the middle of the 21st century, robotics experts hope to create a team of fully autonomous humanoid robot soccer players that can beat the winner of the most recent World Cup.

That’s a pretty ambitious goal, but in the meantime, teams of robots are being used in a broad array of applications from environmental monitoring and mapping to construction inspection and search-and-rescue operations.

Controlling robots in uncertain environments such as over rough terrain or in changing ocean currents is a challenge in itself, but it’s an even greater challenge when the work involves not just one robot but an entire team whose actions need to be coordinated.

Mathematical models can be a valuable tool for controlling robot behavior, but they need to be robust enough to account for the stuff that happens in real life.

To address this problem, a group of researchers at the University of Delaware has created a framework to extend traditional deterministic models, where the output is determined solely by the initial conditions and the applied inputs, into the stochastic regime, where the inherent randomness of the real world can be considered.

In effect, they’ve created a method to predict how well models work.

Their work is reported in a paper, “Probabilistically Valid Stochastic Extensions of Deterministic Models for Systems with Uncertainty,” in the September 2015 issue of TheInternational Journal of Robotics Research, the top journal in the field of robotics.

“Essentially, our method involves the use of data statistics to quantify the amount of uncertainty that the model parameters need to have to capture the variability observed in experimental data,” says lead researcher Ioannis Poulakakis, an assistant professor in the Department of Mechanical Engineering.

“The methodology is general enough to accommodate different robot platforms and types of systems, and it can be applied to a variety of deterministic and stochastic models,” he adds.

To demonstrate the method, the researchers applied it to miniature legged robots that crawl at low speeds as well as to small quadcopters that hover.

The two cases, which involve very different types of uncertainty — leg-ground interaction in the legged vehicle case and aerodynamic effects in the aerial vehicle case — were chosen to demonstrate that there are multiple ways to infuse stochasticity, or uncertainty, into the underlying deterministic model.

“Teams of robots are more fault tolerant than individual robots — if some break down, others can take over,” says Poulakakis. “A team can also reduce the amount of time it takes to complete a given task.”

“We’re hopeful that this method will help us to quantify the effect of uncertainty on individuals so we can build functional robot teams of different morphologies and capabilities for a variety of applications.”

About the research team

The paper was co-authored by Konstantinos Karydis, Ioannis Poulakakis, Jianxin Sun and Herbert G. Tanner.

Konstantinos Karydis earned his doctorate in mechanical engineering at UD in 2015. He is now a postdoctoral researcher in robotics in the University of Pennsylvania’s GRASP Lab.

Ioannis Poulakakis is an assistant professor in mechanical engineering. His research interests are in the area of dynamics and control with application to bio-inspired robotic systems, specifically legged robots, as well as problems pertaining to the dynamics of collective decision making in multi-agent systems.

Jianxin Sun earned his doctorate in mechanical engineering at UD in 2016. He is now with MathWorks in Natick, Massachusetts.

Herbert Tanner is an associate professor in mechanical engineering. His research focuses on multi-robot coordination, motion planning and control, mobile sensor networks, as well as hybrid and cyber-physical system modeling, analysis and design.

Design‘Tiny’ Teamwork