Thesis Contributions


The project focussed on using naïve physics for odour localisation in a cluttered indoor environment. The main contributions of the research are as follows:

 

  1. A new approach to Naïve Physics, encapsulating intuitive rules into an algorithm in order to model an aspect of the environment. This was the first example of Naïve Physics for a practical robotics application.

  2. The identification and analysis of a new problem domain for odour localisation - 
    'enclosed spaces'. Ability to operate in such environments may be essential for many applications such as search and rescue and identification of gas leaks in industrial settings.

  3. A novel, non-reactive method for odour localisation within this domain – modelling the airflow in the environment (using Naïve Physics), and then using this map to reason about odour dispersal, and plan movements that will provide sensor reading that can be used to predict the location of the odour source.

  4. Development of bi-modal search with complementary senses (vision/olfaction) for the localisation of a chemical source. This technique could be applied naturally to many other sensor guided tasks.

  5. A framework for improving the robustness of a Naïve Physics algorithm. Bayesian reasoning on a multiple airflow map hypothesis tree was used to provide robustness against uncertainties in sensor readings, a priori information and errors in the airflow mapping process.