New MIT algorithm helps robots collaborate to get work done
The algorithms allow teams of robots to accomplish missions such as mapping and search and rescue with minimal wasted effort.
In some cases, a robot is not enough.
Consider a search and rescue mission to find a lost hiker in the woods. The rescue team may want to deploy a squad of wheeled robots to scour the forest, perhaps using a drone scanning the scene from above. The advantages of the robotics team are obvious. But coordinating this team is not an easy task. How can I prevent bots from duplicating their efforts or wasting energy on complex research paths?
MIT The researchers designed algorithms to ensure successful cooperation with the team of information-gathering robots. Their approach is based on a balance between trade-offs between the data collected and the energy consumed. This eliminates the possibility that the robot will perform unnecessary operations and get very little information. Researchers say this guarantee is essential for the success of robot teams in complex and unpredictable environments. “Our method is comfortable because we know that it will not fail thanks to the worst performance of the algorithm,” said a doctoral student in the AeroAstro department at MIT. Said Xiaoyi Cai.
The study will be presented at the IEEE International Robotics and Automation Conference in May. Kai is the main author of the treatise. His co-authors include Jonathan How, RC Maclaurin professor of aerospace engineering at MIT. Brent Schlotfeld and George J. Papas of the University of Pennsylvania. Nikolay Atanasov of the University of California, San Diego.
Robot teams often rely on a comprehensive rule to collect information. “There was a guess that gathering more information would never hurt,” Cai says. “If you have some battery life, use it fully and get the most benefits.” Often this goal is implemented in sequence. Each robot assesses the situation and plans its path one after the other. This is a straightforward procedure and generally works well if information is the only purpose. However, problems arise when energy efficiency is a factor.
Kai says the benefits of collecting additional information often diminish over time. For example, if you already have 99 forest photos, it might not be worth sending a robot on a long quest to take the 100th photo. “We want to recognize the trade-off between information and energy,” says Kai. “It’s not always good to have more robots on the move. It can actually get worse, given the energy costs. “
Researchers have developed a robot team planning algorithm that optimizes the balance between energy and information. The “objective function” of the algorithm that determines the value of the task proposed by the robot explains the decrease in the benefits of collecting additional information and the increase in energy costs. Unlike previous planning methods, it does more than just assign tasks to robots in sequence. “It’s a more collaborative effort,” Cai says. “Robots make their own team plans.”
Cai’s method, called local distributed search, is an iterative approach that improves team performance by adding or removing individual robot paths from the overall group plan. First, each robot individually generates a set of potential trajectories that it can follow. Each robot then proposes its trajectory to the other members of the team. The algorithm then accepts or rejects the suggestions of each individual, depending on whether the objective function of the team is increased or decreased. “This allows the robot to plan its own path,” Cai explains. “We only let them negotiate when they need to come up with a team plan, which means it’s a pretty decentralized calculation.”
Distributed local search has proven its spirit in computer simulation. The researchers ran an algorithm against competitors as they tuned out a simulated team of 10 robots. Distributed local search took a while to calculate, but it was guaranteed to complete the robot’s mission with minimal research and keeping team members from getting lost in an unnecessary expedition. .. “It’s a more expensive method,” Kai says. “But performance is improving.”
Jeff Holinger, a roboticist at Oregon State University who was not involved in the research, said the breakthrough could one day help robot teams solve real-world intelligence problems where the energy is a finite resource. They say there is sex. “These techniques can be applied when the robot team has to make a trade-off between the quality of detection and energy consumption, including aerial and ocean surveillance.”
Cai also points to potential applications in mapping and search and rescue (activities that rely on efficient data collection). “Improving this underlying information gathering function is very influential,” he says. The researchers then plan to test the algorithm with a team of robots in the lab, including a combination of drone and wheeled robots.
See also: “Non-monotonic Energy Recognition Information Collection” by Xiaoyi Cai, Brent Schlotfeldt, Kasra Khosoussi, Nikolay Atanasov, George J. Pappas, Jonathan P. How, March 26, 2021 IT> Robotics..
This study was partially funded by the Joint Research Alliance on Decentralized and Joint Boeing and Army Intelligent Systems and Technologies (DCIST CRA).
New MIT algorithm helps robots collaborate to get work done Source link New MIT algorithm helps robots collaborate to get work done