|
| The robot became able to detect the source of ethanol leakage very efficiently and quickly. [Image: Reza Khodayi-mehr] |
Robot nose
Science is far from fully understanding the smell, but electronic noses have tried to imitate this natural sense because it is extremely useful, from the perfume and food industry to the detection of air pollutants and toxic substances.Researchers at Duke University in the United States were just trying to improve the electronic nose created by the team to give a robot the ability to detect the source of the emission of pollutants and toxic leaks.
They quickly discovered that it is not enough to have the robot "follow your nose".
Moreover, although birds such as vultures and vultures can find their source of food tens of miles away, imitation of nature still does not produce good fruit precisely because we do not understand smell.
"Many approaches that employ robots to locate airborne particles are based on bioinspired but simplistic assumptions or heuristic techniques that drive robots against the wind or follow increasing concentrations.These methods can usually locate only one source in open space and can not estimate other equally important parameters such as release rates, "explains Professor Michael Zavlanos.
Physics of airflow and optimized route
In complex environments, current methods can lead robots to areas where concentrations are expected to be larger by the physics of airflows, not because they are the source of the leak.The team found the solution precisely by analyzing these airflows in real-time, which allowed us to track the source of an emission more efficiently.
The robot makes a concentration measurement of the compound in question,
combines it with the previous measurements and solves an optimization problem
to estimate the probability of the location of the emission source. Then it
calculates the most promising place to test this probability and goes there to
carry out the next measurement, repeating the process until the source is
found.
"By combining physics-based models with optimized route planning we can find out where the source is with very few measurements," said Zavlanos. "This is because physics-based models provide correlations between measurements that are not taken into account by purely data-driven approaches, and optimized route planning allows the robot to select those few measurements with the highest information content."
The group is already working to create machine learning algorithms to make their models even more efficient and accurate at the same time. They are also working to extend this idea to programming a fleet of robots to conduct methodical research of a large area. Although it has not yet tested the robotic swarm approach in practice, the team has published simulations that demonstrate its potential.
"By combining physics-based models with optimized route planning we can find out where the source is with very few measurements," said Zavlanos. "This is because physics-based models provide correlations between measurements that are not taken into account by purely data-driven approaches, and optimized route planning allows the robot to select those few measurements with the highest information content."
The group is already working to create machine learning algorithms to make their models even more efficient and accurate at the same time. They are also working to extend this idea to programming a fleet of robots to conduct methodical research of a large area. Although it has not yet tested the robotic swarm approach in practice, the team has published simulations that demonstrate its potential.
|
Bibliography: Model-Based Active Source Identification in Complex Environments Reza Khodayi-mehr, Wilkins Aquino, Michael M. Zavlanos IEEE Transactions on Robotics DOI: 10.1109 / TRO.2019.2894039 https://arxiv.org/abs/1706.01603 |
Tags:
Robotics