Xanity: The Autonomous Savior for Sanitization Workers

In the U.S, 350+ Sanitization workers in New York died in one week due to hazardous working conditions given the pandemic situation. In Canada, sanitization workers in schools have now moved from working every other day to now twice daily to ensure the health and priority of the teachers and students.

However, what about the health of the Sanitization workers? What about those who are risking their lives to clean the environment that could have a high possibility of contamination in order to protect other people.

After conducting a poll in the United States, 80% of sanitization workers felt unsafe about the work that they did every day.

You can’t blame their response. Cleaning the tables, chairs, desks, etc is essential in order to ensure that COVID-19 doesn’t leave its residue for others to unknowingly pick up. For these reasons, the balance between ensuring the safety of people vs that of the sanitization workers is misaligned.

However, this process of sanitizing large indoor buildings and workspaces has the potential to be automated in order to decrease the danger net for the essential worker.

The solution is…


An automated drone sanitization device that can improve both the time and cost efficiency by 3.5x that of manual cleaning. It is also adaptable to several different types of large workspaces and most importantly will save the lives of sanitization workers around the world.

A presented model of Xanity

Xanity works by integrating Computer Vision and Reinforcement Learning combined with several nozzles containing an electrostatic disinfectant spray in order to both process the automation of the drone and simultaneously spray the areas around the environment.

One question arises, will the drone spray everywhere? How will electronic devices be accounted for? What about food, table lamps, and your dog Erelius?

The answer to that is because the chemical composition of the electrostatic disinfectant spray is composed of a harmless solution of water, Peroxide & amp, and 100% natural algae, electronic devices that are going to be sprayed will not be affected. This specific disinfectant was validated by the CBS SF Bay Area company that uses the same electrostatic spray to kill COVID-19 but in the form of a hand-held device.

Now we know that the spray is safe for workplace environments that have several electronic devices. Let’s now dive deeper into how the Computer Vision and Reinforcement Learning algorithm works.

U-Net + Reinforcement Learning

Xanity’s AI model is based on the work of the Computer Engineering department in Sejong University located in Seoul, South Korea. It uses a novel method to perform object detection in real-time in a total of 83ms.

Current collision avoidance AI drones must endure costly manual data labelling because it uses tradition supervised learning models using CNNs. But, Xanity’s model doesn’t purely rely on supervised learning. Instead, it learns to map the environment by itself through Reinforcement Learning, using an action and reward system. Two algorithms will come into play for the RL mechanism to work. The policy gradient algorithm and the actor-critic algorithm.

The Reinforcement Learning process

The policy gradient algorithm is able to remember the history of the past actions that it took from previous states that the drone was placed in. For example, if Xanity turned left at WE Work to avoid a wall, then it will avoid a similar wall at a school, Walmart or other corporate buildings that Xanity has previously visited.

Why is this important?

Well if the drone knows how to move around in a school-based upon its previous actions in WeWork, this means that the drone didn’t have to go through the time-consuming training process.

Secondly, the policy gradient algorithm works in tandem with the actor-critic algorithm. This method allows our RL model to be trained in a virtual environment instead of having to crash all the time in the real world. In the training process, the U-net downsamples the live input image and then uses gradient ascent to optimize the weights. After the training process is finished, Xanity is then deployed into the real environments and uses only the U-net combined with the actor-network to make the most optimal decisions.

Visual Demonstration of the Xanity AI model

Think of the actor as the child and the critic as the mother. In the training process, the actor takes random actions thinking that it’s the best action (i.e eating a plush toy). The critic network (mother) is there to watch over the actor-network and judge its actions. So in essence, in the training process, the critic-network mother raises the actor-network child until the actor-network becomes an adult. Once, it’s an adult, meaning that the weights have been optimized, then Xanity is ready to be deployed into the real world. Once in the real world, the U-net only works with the actor-network now (since it’s an adult and doesn’t need the mother anymore).

Overall, through this process, our model is able to skip the manual data labelling phase, adapt to several different large workspace areas, and most importantly have a reaction time from inputting the image to taking action of only 83ms!

All of this process is able to be done in real-time because of the computationally effective nature of reinforcement learning using non-linear reward models to improve its actions. By having our AI algorithm built through this model, Xanity is able to perform object detection and spray in the necessary areas to sanitize large workspaces.

Alongside the AI component, Robotics is also involved in Xanity’s implementation.


An electrostatic disinfectant sprayer is utilized on Xanity to create a sanitized coating on every surface. An integrated electrode in the spray nozzle gives the atomized disinfectant particles an electro-positive charge. These charged particles attract to almost all surfaces. The particles wrap around all sides of the surface and thereby allow it to be coated with the solution.

This not only enhances sanitization efficiency and effectiveness but also reduces the environmental impact induced by sanitization by eliminating single-use cleaning products. By using a non-corrosive, non-residual and safe disinfectant, the technology ensures no object or human is harmed.

The drone uses an array of cameras and 2D LIDAR Sensors to optimally interpret its environment and any potential obstacles. The onboard Intel Neural Compute Stick enables high-speed data processing and machine learning without the need for a continuous connection with a ground station.

The Lidar sensors, hexacopter model, and spray nozzles depicted in Xanity

Six high power motors enable the drone to carry a payload of almost 10 litres while maintaining a small footprint. Moreover, the high flying speeds allow for a sanitization rate of 1000 sqft/min. Finally, the modular design of the battery pack and disinfectant tank makes for an efficient and simplified refuelling process.

What does Xanity have to offer now and in the future?

With Xanity, sanitation workers won’t have to risk their lives on the front lines against this virus. There will also be no need for PPE equipment and instead, PPE resources can be allocated to the areas that need it most. Because of a drastic cost reduction, Xanity can also make it very cheap to continue with sanitation processes around the globe. Using this autonomous drone, sanitation workers and our world will be safe and healthy.

Although Xanity was created to disinfect against the coronavirus, in the future, it can be used to disinfect hospitals, elderly care homes and more. And if there is ever another pandemic, Xanity will be on the front lines.

A 17 y/o Reinforcement Learning enthusiast looking to leverage tech towards driving social change 🤖

Get the Medium app

A button that says 'Download on the App Store', and if clicked it will lead you to the iOS App store
A button that says 'Get it on, Google Play', and if clicked it will lead you to the Google Play store