Internet of Trays, Your Singapore Hawker Dream

Read on to find out how our IoT solution improves tray return rate at a hawker centre!

11 min readNov 25, 2020

Background

On 16 February 2020, National Environment Agency (NEA), a statutory board under the Ministry of Sustainability and the Environment (MSE) in Singapore, launched the SG Clean campaign. It seeks to rally members of the public to step up on personal hygiene in order to raise standards of cleanliness in Singapore and safeguard public health.

Despite campaigning efforts, the percentage of patrons who return their trays and crockeries remains low. With manual surveys conducted at Berseh Food Centre (FC) documenting tray return rates as low as 20–30% compared to newly launched hawker centres, hovering around 60%.

Facing difficulties with collecting accurate tray return data, it has been a daunting challenge for NEA to measure the effectiveness of their tray return campaign. Much less identifying possible ways to improve their current campaign. Finding a solution would allow NEA to achieve the following strategic objectives:

Better hygiene conditions as the hawker centre is clean, also allows patrons to find seats faster
Able to evaluate the effectiveness of existing solutions and pinpoint areas of improvement
In line with SG Clean goal — to raise standards of cleanliness and public hygiene in Singapore and safeguard public health

Thus, in collaboration with NEA, our team, Internet of Trays from Singapore Management University, decided to leverage on the potential of Internet of Things (IoT) to improve their data collection process.

Are existing solutions sufficient?

What has MSE done so far?

Manual Data Collection
Berseh FC’s current tray return rates are measured by allocating a manager to manually oversee a zone of roughly 20 tables. By recording the behaviour of patrons from the first 2 tables who have completed their meals, the manager then observes the total number of trays returned out of the total trays used and takes that as the zone’s tray return rate.

Tray Return Campaign
To improve on the tray return rate, NEA has spread tray return campaign posters around the hawker centre to encourage and remind patrons to return their trays.

Also, tray return racks have been strategically placed around the hawker centre where it is convenient for patrons to return their trays to encourage patrons to do their part.

Tray Return Rack strategically placed nearby tables

Who are the stakeholders affected?

NEA
This tiresome process of measuring the tray return rate does not truly reflect the true tray return situation in Berseh FC due to human error and inefficiency in data collection. Thus, NEA may gather inaccurate tray return rate data resulting in non-actionable insights.

Furthermore, it fails to measure the effectiveness of tray return campaigns to provide insights as to how they might be able to improve the campaign.

Customers
The lack of data required to improve tray return rate would continue to result in poor hygiene conditions at the hawker centre as well as making it difficult for customers to find a table. These would give customers a negative experience when dining at the hawker centre.

Introducing Internet of Trays

So what makes our solution so unique?

Market solutions out there generally use RFID technology in order to gather tray return data. The nature of RFID allows for quick detection with the ability to take multiple readings at once so it doesn’t matter even if the trays are stacked.

However, not implementing the RFID system hawker-wide greatly dampens its effectiveness. Take for example, the following scenarios:

Not returning the trays to their designated RFID point would lead to underestimation of tray return rates
The possibility of wear and tear or misplacing our RFID tray tags from washing
Lack of resources to tag sufficient trays for data collection

Despite being a great solution, it fails to account for many blind spots thus we decided to proceed with a zone-based solution that would provide a better reflection of the tray return rate for a selected zone given the known constraints.

Our solution is founded on the following principles we have learnt in IoT:

Minimal deployment of sensors, use only what is needed.
Easy to deploy and easily scalable.
Relatively friendly and appealing aesthetics.

Architecture

Here is a high-level overview of our implementation, our group makes use of 3 different types of sensors in total. The RPI acts as a gateway to send data to our Mongo Database through WiFi connection. This data is then constantly being aggregated by an external EC2 server running on AWS which is then pulled by our Frontend Dashboard to display up-to-date information online.

Our Solution Set Up

Tray Return Rack
This setup comprises 2 Infrared (IR) sensors on each level of the Tray Return Rack.

By emitting an infrared light which returns the distance of the closest object obstructing its path, the IR sensor is able determine whenever a tray is present. By keeping track of the status of the tray return rack we would be able to track its utilization rate.

Table Set Up
As for the table’s setup, we have used 2 different sensors to ensure accurate data collection.

The round force sensors (RFS) informs us when a tray is placed on or removed from the table by detecting the change in pressure. While, the Proximity Infrared (PIR) sensor detects movement thus allowing us to check if the table is occupied. This allows us to see if patrons have left without returning their trays, essentially reflecting the tray return rate of a given table.

These have all been taped down and hidden from the customers making them as unobtrusive as possible. Above the RFS, we also placed boards with poster cut outs of the tray return campaign to encourage customers to return their trays.

Charts of Wisdom

The data received is displayed on our live dashboard so we can observe real-time changes in occupancy status of the tables and utilization of the tray return racks.

Moreover, there is an insights tab that automatically generates insights based on the data acquired which can provide the management with a direction to improving the tray return campaign. Read on to the next section to find out more!

Satisfying our stakeholders

NEA
With a frontend dashboard that visualises the data collected, NEA will be able to identify correlations that are attributed to the tray return rate. This provides NEA with actionable insights to improve the tray return campaign.
Customers
With an improved and effective tray return campaign, customers will be more likely to return their trays. Thus, more tables will be cleared and customers will be able to look for a seat quickly and efficiently, especially during peak periods.
Management
More efficient allocation of manpower to ensure the cleanliness of the hawker centre.

Discovery is the Journey

We deployed our IoT solution for a day on one Table (Table1) and one Tray Return Rack (TRR1). Through which, we managed to gather interesting insights about tray return rate and trr utilization within Berseh that we want to share.

A peek at some of our insights!

1)Tray Return Rate

By aggregating the data we collected with the measurement of crowd levels that we pulled from Google API, we found that tray return rates are generally higher during non-peak periods.

2) Average Clear Time for Tray Return Rack

Figure 1.2. Average Clear Time of Tray Return Rack Against Crowd Level Visualization

By visualizing the average clear time of tray return rack against crowd level, we looked to explore factors which could potentially affect the tray return rate. Such as whether the tray return rack is overcrowded and neglected. Although there were times where the cleaner took a longer time to clear the rack, it was never full. This means that the rack has not been fully utilised.

3) Average Clear Time of Tables by Cleaners

Figure 1.3. Average Cleaner Response Time to Clear Tables Against Crowd Level

From this graph, we saw that there is a generally higher response time for cleaners to clean the tables as compared to tray return racks (refer to figure 1.2) especially during the peak periods. Evidently, the larger spread of tables across Berseh makes it harder for cleaners to get to them.

Having analysed the data, NEA would be able to:

Improve tray return rate by putting up additional posters within the proximity of Table1
Increase utilisation of trr1 by relocating it closer to tables to make returning of trays more convenient for patrons, which in turn increases tray return rate.
Reallocate Manpower to deal with varying crowd levels.

Daily Overall Insights based on Tray Return Rate data

Given the success of our proof-of-concept solution, we believe that by increasing the scale of implementation, we would be able to strengthen the tray return campaign at potentially challenging zones by identifying:

Tables that contributed the most to tray return rate to identify attributes that would lead to an effective tray return campaign e.g. ideal distance of table from poster.
Tables with the lowest tray return rate which may require the tray return campaign in those areas to be improved.

#ChallengeAccepted

What are the challenges that we faced?

Ensuring data accuracy was the main challenge given the unpredictability of real-world scenarios that we were not able to account for.

Improving Data Accuracy
One critical concern we found through experimentation was that being still for too long while having our meals tricks the motion sensor into thinking the patron has left. The poor sensitivity of the motion sensor meant that we were unable to rely on the change in binary values to detect an “unreturned“ tray. To reduce the risk of false readings, we programmed the sensors to only register a tray as unreturned if the table is unoccupied for 10 seconds or more.

Even then, the sensitivity of our sensors remains problematic. Our group had initially placed our force resistors underneath a sticker so they appeared less intrusive. But we later found that the sticker was weighing down on the sensitive force resistors, thus leading to an inaccurate reading.

Our team’s workaround was to place a corrugated board over it instead. Although this has improved data accuracy, there was a tradeoff in the form of intrusiveness to patrons.

For future iterations, the force resistors can be calibrated or have its sensitivity adjusted using a potentiometer. This not only improves data accuracy but also ensures a non-intrusive design.

In addition, the sensitivity of our PIR sensors often led to inaccurate readings of table occupancy that arose from passers-by walking past the table. Seeing that, we have taped it up to limit its range so that it only focuses on the area specifically underneath the table.

PIR sensor taped to increase data accuracy

Growth of Reality

What are some limitations that we faced?

Limited Budget
Due to our limited budget, we were unable to procure more sensors to measure the tray return rate across different zones — a practice by which NEA has been adhering to. However, our solution allows us to measure the effectiveness of the tray return campaign and tray return rate of a designated zone, and can be easily scaled up to include other zones.

Limited Data
As Berseh FC is a public space, there was a risk of tampering from the public, thus requiring our physical presence in collecting data. Moreover, there is only a small window of availability to deploy our sensors in order to minimize disturbances to patrons. Given these circumstances, the amount of data that we could collect was severely impaired as our schedule only permitted us to be on the ground for a day. This has in turn made it difficult to affirm the correlation between tray return rate with crowd level or draw much deeper insights for that matter.

Key Takeaways

What are some learning points that we’ve gained?

Feedback

“Despite being somewhat intrusive, this solution is highly detailed and encompassing” — Feedback by MSE

Our team has successfully collected tray return rate, utilization of tray return rack as well as other factors that could affect tray return rate. Reflecting these insights in a form of a dashboard and charts, we managed to provide actionable insights for NEA to improve their tray return campaign. Despite a highly detailed solution, the design of the solution can be improved to ensure a less intrusive solution as mentioned by NEA.

Final Thoughts

“Our IoT journey was a tiring but rewarding one.” — Team Internet of Trays

Seeing how we turned our ideas to a working prototype, then proceeding to implement them at an actual site was a sense of achievement for us as we’ve witnessed how our solution can solve a real-world problem.

We’ve learnt the importance of having foresight as we would have to anticipate different behaviours that the patrons have to refine our solution continuously to ensure that the data collected is accurate. Every solution is bound to have its limitations and constraints that we have to overcome and we believe that this project has provided us with the experience that we can bring forward to our future projects!

Let’s show some appreciation

We would like to thank Prof. Hwee Pink and Pius for guiding us to ensure that our solution is implementable. Not forgetting our TA, Vinnie, we thank you for your hard work in coordinating and ensuring a smooth communication between us and our project sponsor!