The upper photos is a common scene happens at Columbia every day: people are waiting in long lines to get lunch and dinner. It can get extremely frustrating when people have to do this in cold winter nights and in rainy or even snowy days (you can see an umbrella in the last photo). Last semester, we managed to give a solution to this problem with an initial prototype of Smart Food Delivery system. However, our solution had not taken usability, reliability and environmental impact into serious consideration. This semester, our group presented an improved prototype, which focused more on data analysis instead of hardware construction. It managed to save heating energy and provided more interaction with users as well as solving the food waiting problem.
The first version of our prototype works as a "heated box" with lighting and QR code recognising functions. It serves as a relay between the food delivery guy and user. This can prevent users from wasting time in lines because picking food up from our boxes is significantly faster then waiting for food being cooked in a food track. However, the first version of our prototype overlooked heating energy used and lacked user interaction, and these problem were left to be solved this semester. Another challenge we faced was the previous database we used (Parse) is being stopped this year, so we decided to migrate our database to AWS DynamoDB. Also, the boxes we used for the first version of our prototype was too small to be functional, so we rebuilt the hardware system with new boxes and a different sensor system, with the developer kit provided by the labs this semester. Furthermore, We developed box cleaning alert function to improve user experience. We also use push notifications to inform restaurants and users about actions required and order status.
In the prototype we built last semester, we inform the restaurant about a new order immediately after the order is placed. If a user placed his lunch order at 8:00 in the morning for his own convenience, the food can be kept in the box heating for 4 hours before the user come and fetch it. This can be very energy consuming and may impact the taste of the food. We solved the problem by predicting the time spent on cooking and delivering for every order with a machine learning model, and inform the restaurant in a "perfect" time to shorten the storage duration. For instance, for the previous example, although the user placed order at 8:00, our system will inform the restaurant about the user at 11:30, so the food can arrive at the box not too long before the user's fetching time. We also applied machine learning for the cleaning function. By measuring the humidity difference inside and outside the box for a period of time, we will apply our classification model to decide if there is spill in the box, and therefore if the box needs cleaning. If the box do needs cleaning, we will inform a cleaner to come and clean the box.
The first version of our prototype works as a "heated box" with lighting and QR code recognising functions. It serves as a relay between the food delivery guy and user. This can prevent users from wasting time in lines because picking food up from our boxes is significantly faster then waiting for food being cooked in a food track. However, the first version of our prototype overlooked heating energy used and lacked user interaction, and these problem were left to be solved this semester. Another challenge we faced was the previous database we used (Parse) is being stopped this year, so we decided to migrate our database to AWS DynamoDB. Also, the boxes we used for the first version of our prototype was too small to be functional, so we rebuilt the hardware system with new boxes and a different sensor system, with the developer kit provided by the labs this semester. Furthermore, We developed box cleaning alert function to improve user experience. We also use push notifications to inform restaurants and users about actions required and order status.
In the prototype we built last semester, we inform the restaurant about a new order immediately after the order is placed. If a user placed his lunch order at 8:00 in the morning for his own convenience, the food can be kept in the box heating for 4 hours before the user come and fetch it. This can be very energy consuming and may impact the taste of the food. We solved the problem by predicting the time spent on cooking and delivering for every order with a machine learning model, and inform the restaurant in a "perfect" time to shorten the storage duration. For instance, for the previous example, although the user placed order at 8:00, our system will inform the restaurant about the user at 11:30, so the food can arrive at the box not too long before the user's fetching time. We also applied machine learning for the cleaning function. By measuring the humidity difference inside and outside the box for a period of time, we will apply our classification model to decide if there is spill in the box, and therefore if the box needs cleaning. If the box do needs cleaning, we will inform a cleaner to come and clean the box.