The Synthetic Moth project has been developed as a part of the Neurochem european project, funded by the European Commission under FP7, Bio-ICT Convergence – FET programme.

The work presented here was published in the 2010 IEEE World Congress of Computational Intelligence. The complete citation of this work can be found in Publications.

Autonomous robotic odour source classification and localization in real world environments is an essential step for applications such as humanitarian demining, environmental monitoring or search and rescue operations. However, at the moment, this problem has only been solved by nature (e.g.: moths, bees, rats, dogs). Biological systems are capable and efficient at odour source localization in spite of the difficulties present in the real world such as turbulent environments, obstacles, predators or interfering odours. Here we aim at exploiting our understanding of the moth to solve this problem and we propose a biologically based model of moth behaviour. We implement our model on a robot that uses chemical sensors and we test its performance in a controlled environment. Further, we extend the behavioural model with a sensor front end that supports classification in order to deal with odour distractors. We show that our system is able to locate an odour source and map the chemical environment in the presence of distractors.

Our model combines chemical information (chemotaxis) and wind information (anemotaxis) to perform the search tasks. It is based on the well known search behaviour of the male moth when the female releases pheromones in a strong airflow dominated environment generating a plume dispersion downwind. Our search localization model has two different modes: surge and casting. The surge mode makes the robot move upwind when it detects odour filaments in the plume. If it looses the plume, the casting mode is enabled which generates an increasing zigzag trajectory until it reencounters the plume. This process is repeated until the robot finds the source of the odour or leaves the test environment.

The model has been implemented using iqr, a multilevel neuronal simulation environment that provides a tool for graphically designing large-scale real-time neuronal models and allows to test them in real-time using a real world systems such as robots. iqr runs on a  desktop computer and it is interfaced to the robot via a bluetooth connection. It acquires data from the robot sensors, processes it and sends motor commands at run time. The vision based tracking system sends the data to iqr via a direct UDP connection.

IQR screenshot: neuronal model example

The robot we use for the experiments is called SPECS M2. Since the robot does not have a wind direction sensor and we generate a strong airflow of known direction inside the wind tunnel, we use the tracking information of the robot to simulate the wind direction sensor. The odour blend is measured by means of chemical sensors that are placed at front of the the robot and connected to its main board.

The artificial moth: SPECS M2 Robot

The experiments were carried out in a wind tunnel. This is built of wood and covered with transparent low density polyethylene, divided in three modules. The first one comprises four exhaust ventilators that generate a wind flow inside the tunnel. The other two modules form the tunnel and create a controlled space where the robot can move freely. The vision based tracking system (AnTS) is placed above the tunnel. To disperse the compound we use an ultrasonic source that generates a mist at the entrance of the wind tunnel (for clarification purposes, the entrance or the beginning of the wind tunnel is where the odour source is placed and the exit or the end where the fans are located). Then the negative pressure created by the four fans creates a plume that moves across the whole wind tunnel from the entrance to the exit where the air is extracted out of the experimental environment.

Wind tunnel

Results show that the chemical sensors and can be used to reconstruct the plume in a way that allows for source localization. The performance of the casting and search behavioural model in 17 odour source localization experiments showed high performance. Moreover, the results also show successful odour classification when the robot explores the tunnel in presence of different substances.

Plume reconstruction performed by the robot

Spatial classification of two different chemicals

Publication:
J.Blanco Calvo, Sergi Bermúdez I Badia, Héctor Tapia Simón, Paul F.M.J. Verschure. The real-world localization and classification of multiple odours using a biologically based neurobotics approach. 2010 International Joint Conference on Neural Networks, IEEE. Barcelona, 2010.

Credits:
José M. Blanco Calvo, Sergi Bermúdez I Badia, Héctor Tapia Simón, Paul F.M.J. Verschure.

Thanks for the collaboration to Zenon Mathews, Miguel Lechón, Ulysses Bernardet and Aleksander Valjamae.

Special thanks to Sergi Bermúdez for his continous support and guidance. Also I want to thank Paul Verschure for the opportunity of carrying out this work in SPECS.