Team name: Mozzie Cloud Shaker
Team member’s full names: 1- Syed Qaisar Jalil 2- Stephan Chalup 3- Jamil Khan 4- Soultana Papasotiriou and prossibly additional experts
Team Lead’s name: Syed Qaisar Jalil
Name of Project with a one-sentence description An attractor and repeller framework of intelligent networked mozzie traps using multi-agent reinforcement learning and communication network techniques to make an outside area mozzie free
Create a link from YouTube to a video of no more than 3-minutes duration that outlines your project and its proposed outcomes.
A longer description of what your Project is about, including these headings (no more than 1,000 words in total):
Define the problem you are seeking to solve. During summer at UoN it becomes impossible to sit outside, e.g., the center courtyard in the Union building, due to the mozzies. Our aim is to propose a solution that will make the outside sitting places mozzies free.
Describe your big idea (what is the vision?). The idea is to create intelligent attractor and repeller using multi-agent deep reinforcement learning (which is a branch of machine learning). These multi-agents will be connected via a wireless network.
Describe and illustrate (if possible) your proposed solution. The approach we are suggesting is based on multiagent reinforcement learning.
It will be used to estimate and control (“shake”) a cloud of insects by a ring of agents. The agents are networked CO2 emitters, light emitters, fans and sensor-traps that can count sapped animals.
I.e. we do not trace Mozzie flight paths but only the endpoints and timing when they enter the traps. The zap count over time is the reinforcement learning signal required to train our multiagent system.
As there would be a ring of sensors (e.g. 10) around an area like the center courtyard in the Union building the zap signals will provide some information about the number and dynamics of the insect cloud.
If we can control CO2 emitters, fans and/or light at the (stationary) sensor-traps, they act like 10 trained shepherd dogs “running around” or better, standing around and barking at a herd of cattle (the Mozzies).
- What is the core foundation of your research or solution (this could be technology, research, know-how, etc)? The core foundation of our project is the use of a multi-agent system which is trained via reinforcement learning.
We also assume that mozzies swarm or somehow interact. This is still an open research question according to the challenge's expert panel.
Describe who you think your end-user and/or paying customer could be. The end product can be used for any sitting place such as sitting places in university, hospital, etc.
Describe your Technology Readiness Level or Research Literature Level. The project development is yet to be started.
Describe the top three critical hypotheses you want to explore, including: How you will test them; Describe your experimental plan, including any new technologies or tools to be developed; and If your experiment/s in the testing phase is successful what are the next steps?
Initially, a simulation study should reveal how much training time is required to establish this optimal mosquito trap.
The trap circle would also be useful and effective without training and could be sold.
If a few hundred of them are installed they can then all be networked and this way shorten training time and jointly achieve optimal behavior. The required scale and feasibility of the approach could also be tested in simulation first.
The system can also include cameras to check where humans or animals are in the scene and what the lighting is like.
We will also develop a hardware prototype of the mozzie trap with mozzie counter.
Short summary: Essentially we will modify existing zapper traps by adding a counter (how many mozzies are zapped) and we will network and control them. This should make the group of traps to work together as a team and make them more efficient.
- Describe how would use the funding to progress your hypotheses, including:
How will the work described be performed within the budget (up to AU$5,000) and time period (6 months) allocated for the testing phase (resources, capability, etc)? What essential outcomes will you generate during your testing phase? Include a brief breakdown of allowable costs.
$2500: Development of a simulation of a mozzie swarm and how it is controlled by a ring of intelligent networked mozzie traps with counter. The main techniques is multi-agent reinforcement learning.
$2500: Hardware prototype of modified networked mozzie trap with counter.
- What you’ve done to date, including challenges and wins.
We have experience with reinforcement learning, hardware development, networking and simulations.
- Why your idea is an unconventional or creative approach to the problem. The recent advances in the field multi-agent reinforcement learning have allowed obtaining an optimal policy in learning problems. Now, formulating the mozzie problem as a learning problem where multiple agents are connected and working together to make the place mozzie-free is an unconventional approach compared to the traditional methods where multiple devices are deployed without any communication or coordinated behavior.