When a robot sets out to reach a destination, the decision-making process can vary significantly based on the complexity of the environment. If faced with only two routes, the robot can easily compare travel time and success probability. But, as the number of potential paths increases, determining the best option amid uncertainties can quickly become overwhelming.
In response to this challenge, researchers at MIT have devised an innovative algorithm that aids robots in effectively navigating complex environments. This method constructs roadmaps that strike an optimal balance between the quality of the routes and computational efficiency, allowing robots to identify the quickest and most reliable paths.
The algorithm commences with safe pathways and intelligently identifies potential shortcuts, reducing overall travel times without sacrificing reliability. In various simulated scenarios, the team demonstrated that their algorithm consistently outperformed traditional models that prioritize either efficiency or performance.
This groundbreaking work has extensive potential applications, such as guiding robots on exploration missions across the rugged terrain of Mars or enabling search-and-rescue drones to locate individuals stranded in remote areas like mountainsides.
“In expansive outdoor settings, it isn’t feasible to know with certainty which paths are traversable. However, with even minimal information about the environment, we can create a high-quality roadmap,” explains Yasmin Veys, a graduate student in electrical engineering and computer science (EECS) and the lead author of a recent paper discussing this technique.
Co-written by Martina Stadler Kurtz from the MIT Department of Aeronautics and Astronautics, and senior author Nicholas Roy, a professor in the same field connected with the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL), this research will be showcased at the International Conference on Robotics and Automation.
Graph Generation in Robotics
Researchers often portray a robot’s operational environment using a graph structure, where “edges” represent possible paths between its origin and destination. The team utilized a particular graph model known as the Canadian Traveler’s Problem (CTP), a concept inspired by Canadian drivers rerouting due to snow-blocked roads.
In a CTP, each edge features a weight indicating the time required to traverse that path, along with a probability rating reflecting its potential traversability. The primary objective is to reduce travel time while maximizing route reliability.
The researchers concentrated on the automatic generation of a CTP graph that accurately depicts an uncertain environment. “Navigating without any information can be daunting. By utilizing partial data—like satellite images—we can create a better representation of the environment,” comments Kurtz.
This algorithm accounts for partial information, categorizing areas (like lakes or fields) and assigning probabilities to their traversability. For instance, non-aquatic robots have a higher chance of successfully crossing open fields than lakes.
With this data, the algorithm initially builds a conservative graph of safe spaces, identifying a slower but reliable route. Then, it introduces potential shortcut paths through uncertain areas that could reduce travel time, guided by an innovative evaluation metric.
Optimizing Route Selection
By focusing only on shortcuts deemed likely to be traversable, the algorithm simplifies the planning process and avoids unnecessary complications.
“The effectiveness of motion planning hinges on the quality of the graph. If this graph lacks quality paths, a dependable plan can’t be generated,” Veys elaborates.
In testing over 100 simulations across increasingly complex settings, the researchers consistently found that their approach surpassed baseline techniques that fail to consider probabilistic factors. They also demonstrated effectiveness in practical settings using a mapped aerial view of the MIT campus.
Looking ahead, the team aims to enhance their algorithm for multidimensional applications, which could significantly broaden its use for intricate robotic tasks. Additionally, they seek to explore the differences between CTP graphs and the real-world environments they model.
“Robots working in real-world scenarios face numerous uncertainties due to varying sensor data, the unpredictability of environments, and the behavior of other agents. Addressing these uncertainties often leads to significant computational demands,” says Seth Hutchinson, a Robotics professor at Georgia Tech who was not involved in this research. “This study offers a smart approximation strategy to efficiently generate uncertainty-tolerant plans.”
This research received partial funding from the U.S. Army Research Labs under the Distributed Collaborative Intelligent Systems and Technologies Collaborative Research Alliance, as well as the Joseph T. Corso and Lily Corso Graduate Fellowship.
Photo credit & article inspired by: Massachusetts Institute of Technology
