Robots are doing more in the air, on the ground and in the water -- but sometimes they get confused and their systems break down. That's why researchers at MIT are developing a system that allows robots -- or other agents in distributed systems -- to make better decisions when there is no clear right answer.
For the past year, researchers from MIT's Computer Science and Artificial Intelligence Laboratory (CSAIL) have been working with robots and developing the model and algorithms for a new system that allows robots, cameras, humans and other agents to work together in scenarios where the outcome or environment is uncertain. Most of their research so far has been focused on warehousing environments, said Post-Doctorate Associate Christopher Amato, but the system is applicable to a wide range of applications including firefighting, search and rescue, or any other scenario that includes a heterogeneous pool of agents and an environment filled with uncertainties.
Robots, drones and sensors are becoming more common, and researchers like Amato and his team are solving one of the biggest problems with those technologies: that things often go wrong, and human intervention is needed to do the thinking that fixes the problem. There are mechanical failures, communication imperfections, sensor failures and mixed signals. The high-level decision-making that humans can employ is often absent from distributed systems of robots and other sensors.
Their research is far from complete, Amato said, but they recently had a breakthrough, which was getting their theory and algorithms to scale to a larger group of agents.
“We have a mathematical model that can model almost any situation where there are multiple agents involved – people, sensors, cameras, robots, whatever,” he said. “But since it’s so general, it’s really hard to solve, which means it’s really hard to generate solutions for what the best thing to do is in all those situations. Our first breakthrough was to come up with a way that is more scalable so we can solve problems of interest.”
In warehouses like the ones run by companies like Amazon, for example, ground robots are used to retrieve pallets of goods, and those robots work together with people who pack the goods into boxes and handle other administrative tasks. The CSAIL researchers are looking at warehousing as an example of an environment where those agents work together to reach a goal and overcome uncertainty in all its various forms. The problem, Amato said, is that current distributed systems assume perfect information, but the world isn’t perfect.
MIT Research in Action
Take a peek at research from MIT's Computer Science and Artificial Intelligence Laboratory on developing the model and algorithms for a new system that allows robots, cameras, humans and other agents to work together in scenarios where the outcome or environment is uncertain.
“Current systems now do assume that robots know where every box is, every second of time,” he said. “There are lots of other settings where that’s not the case, and one of them is search and rescue. You don’t know where people are – you just need to rescue them.”
Their system assumes that there will be communication problems, that there are unknown objects and situations that could arise in the environment, and the theoretical model and algorithms they’ve developed are intended to work around those issues and reach whatever the goal is in the most efficient fashion possible.
The team’s research is available now for any technology company or roboticists looking for a better model to govern their distributed network of heterogeneous agents, and the team may also release their source code at some point, too, Amato said. How far their research will continue will depend on the funding that is available, but Amato said this is the type of work he wants to do, and he will continue striving for improvements in this field throughout his career.
The team, which consists of Amato, George D. Konidaris, Gabriel Cruz, Christopher A. Maynor, Jonathan P. How and Leslie P. Kaelbling, will present its research at the International Conference on Autonomous Agents and Multiagent Systems in May.
Colin wrote for Government Technology from 2010 through most of 2016.