Estimating exposures from spray products using robotic simulations

Abstract

There is an increasing need for exposure data to enable more precise information for risk estimates and improved public health protection. While personal monitoring data are preferred, it is often difficult to collect due to the resources needed to complete a human research study. In this study, we successfully programmed a robotic arm to mimic human use (spraying) of a fabric crafts protector (FCP) and human cleaning (spraying and wiping) of a glass pane with glass cleaner (GC). The robot was then used in place of human subjects to assess inhalation exposures to volatile organic compounds (VOCs) during the use of the FCP and GC. Air sampling data were collected while the robot used the products to estimate personal exposures to VOCs. Average VOC concentrations were 1.57 ppm for FCP spraying and 0.17 ppm for GC spraying and wiping. During FCP spraying, average acetone concentrations were 0.88 ppm and average isopropyl alcohol concentrations were 0.26 ppm. During GC spraying and wiping, average 2-butoxyethanol concentrations were 0.15 ppm. Air sampling data were found to be within the range of data reported in the literature during human use of similar glass cleaning products. No data was found in the literature during use of fabric protector spray products. This study contributes exposure measurement data with detailed contextual information to help characterize inhalation exposures during the use of 2 spray products. In addition, the study offers a systematic, efficient method for generating exposure data which can be used to improve health and safety risk assessments used for public health protection.

What’s Important About This Paper?

This study demonstrates the use of a robotic exposure simulation system to assess exposures to spray products. A robotic system offers an efficient method for the collection of exposure data. Such data can be utilized with exposure models and as part of a comprehensive exposure assessment strategy.

INTRODUCTION

Exposure assessment is a process of estimating the amount, frequency, and duration of exposure to a substance and is one of the core components of risk assessment and risk management (National Research Council, 2012). Ideally, personal inhalation exposure measurements or air concentration data collected in an individual’s breathing zone are collected using sampling techniques that record an individual’s direct exposures during a task or activity for a specified time period.

Occupational hygienists have opportunities to collect personal air sampling data on workers in the workplace. However, collecting exposure data on consumers and for infrequent worker tasks (e.g. tasks performed once a year or less) is more limited and often requires additional time and resources. For example, researchers may conduct human studies where volunteers simulate handling of materials, or where consumers are observed in their everyday lives using the products and chemicals of interest. These types of studies require time to obtain ethical approvals and to recruit and train volunteers, leading to less exposure measurement data collection and more reliance on exposure models. Though exposure models are useful tools for estimating exposures, measurement data are necessary for proper model validation and refinement (Wouter 2017).

The use of robots offers an alternative approach to collecting exposure data, eliminating the majority of aforementioned challenges while also providing an opportunity to systematically evaluate exposure modifying factors. Instead of human subjects, robots or any machine capable of simulating human-like motions can be used to generate air concentrations of chemicals which can be collected and used to estimate personal exposures. The robot can repeat the same motions allowing for evaluation of exposure modifying factors without concerns related to human variability; on the other hand, the programming can be intentionally modified with all other experimental conditions kept the same to systematically assess the impact of human variability. Removing the need for human subjects minimizes ethical considerations related to intentional exposure and eliminates time needed to recruit and train volunteers, which can enable timely data collection. Lastly, the hardware, software, and programming used to simulate human motions in 1 study can be used by other researchers to generate data needed to evaluate exposures during similar activities. This can provide both efficiencies in data collection and continued validation and refinement of exposure models.

A proof of concept study completed by Feld-Cook et al. (2021) demonstrated that a robot could be used to estimate inhalation exposures during drywall painting using a roller application. The goal of this study is to further test the feasibility of using robots to mimic human use of spraying products and to assess the reliability of using exposure measurement data generated during robotic simulations to characterize human inhalation exposure to chemicals during spraying tasks by comparing robot-generated data to human exposure data published in peer-reviewed literature.

Materials and methods

Two commercially available consumer spray products were selected for this study: a fabric crafts protector (FCP) and a foaming glass cleaner (GC). Manufacturer instructions were used to initially characterize the motions needed for the study to inform selection of an appropriate robot. To mimic human use of the products as closely as possible, a human volunteer’s motions were mapped using a 3D camera and tracking markers. This human motion data was then used to generate the motion algorithms for the robotic arm. More details on the robotic hardware and software selected, designed, and used for this study can be found in Supplementary Information 3.

Robotic simulation of spray product use

Task 1. FCP spraying

To simulate human use of the FCP, a robot was programmed to spray FCP downwards onto 2 pieces of 12 in × 8 in craft felt, which were laid side-by-side inside a 22 in (left to right) × 16 in (depth) × 15 in (top to bottom) cardboard box to capture overspray (Fig. 1a). The robot shook the can 4 times and held the spray trigger releasing product for approximately 11 s while moving its arm in a zig-zag pattern from top to bottom and back to the top covering both pieces of felt with product.

Fig. 1.

a) Fabric crafts protector spraying. b) GC spraying and wiping angle 1. c) GC spraying and wiping angle 2.

Task 2. GC spraying and wiping

To mimic human use of the GC, a robot was programmed to spray and clean an 18 in × 24 in clean glass pane. The glass pane was placed on top of a sheet of heavy-duty craft paper inside a large cardboard box, approximately 22 in (top to bottom) × 36 in (left to right) × 16 in (depth) to capture potential overspray (Fig. 1b,c). The robot shook the can 4 times and sprayed the GC downwards onto the glass pane in a zig-zag pattern from top to bottom (product released for approximately 6 s). The robot then wiped the pane clean with a clean and dry microfiber cloth in a zig-zag pattern from top to bottom and back to the top (approximately 30 s).

Exposure data collection

Safety data sheets (SDS) were used to identify the chemicals to assess for each product (see Table 1). Both tasks were performed under 2 conditions, with local exhaust ventilation (LEV) and without LEV, to demonstrate that the hardware and software could be used to systematically evaluate the impact of exposure-modifying factors such as ventilation. The FCP spraying task was performed 3 times with LEV and 3 times without LEV for a total of 6 trials during which acetone, isoprpopyl alcohol(IPA), and total volatile organic compounds (VOCs) were measured in air (a trial refers to robotic simulation of using the product once). For all 6 FCP spraying trials, the robot performed the same shaking and spraying motion once. The GC spraying and wiping task was completed 5 times with LEV and 5 times without LEV, for a total of 10 trials during which 2-Butoxyethanol (2-BE) and total VOCs were measured in air. For the first 2 trials, the robot wiped the glass 2 times; trace amounts of foam were observed on the glass pane, so the programming was updated to have the robot repeat the wiping motions 3 times for the remaining 8 trials.

Table 1.

Composition of FCP and GC per manufacturer SDS.

	Ingredient	CAS	% by weight
Fabric crafts protector	Acetone	67-64-1	37–41
	Isopropyl alcohol	67-63-0	31–35
	Light alkylate petroleum naphtha	64741-66-8	17–21
	Carbon dioxide	124-38-9	2–6
	Fluorochemical urethane	–	<3
Foaming GC	2-Butoxyethanol	111-76-2	2.5–10
	Ethyl alcohol	64-17-5	2.5–10
	Butane	106-97-8	1–2.5
	Propane	74-98-6	1–2.5
	Other components below reportable levels		90–100

Sample collection

All trials were completed inside the University of Texas at Austin Human-Centered Robotics Laboratory (UT Austin HCRL) in October and November of 2020. During each trial, air samples were collected using a combination of photoionization devices (PIDs), whole air sampling canisters, and charcoal tubes as described below. Air sampling equipment was located on a fixed platform about a human arm’s length from the spray can (22 inches) to simulate breathing zone sampling. The actual air intake for the samples varied depending on the equipment. For example, for the FCP spraying task, the minicans needed to stand upright with no additional tubing; thus the air intakes were approximately an additional 6 inches away from the spray can. On the other hand, the PIDs could be laid down flat and off the edge of the fixed platform with the air inlet closer to the spraying which would have pulled in air about 4 to 6 inches closer to the spray can. For the glass cleaning task, the pump and tube samples were elevated due to the height of the pump (approximately 5 inches) but could be positioned closer to the spray can with the flexible tubing; and again, the PIDs could be laid down flat and off the edge of the fixed platform with the air inlet closer to the spraying (see Fig. 1c). All air samples were collected in duplicate as follows: during each FCP spraying trial (n = 6), 2 PIDs were used for continuous volatile organic compounds (VOC) measurements and 2 whole air canisters were used side-by-side to collect integrated acetone and IPA concentrations for a total of 12 VOC, 12 acetone, and 12 IPA samples. During GC spraying and wiping, 2 PIDS were used for continuous VOC measurements during all 10 trials and 2 charcoal tube and pump samples were used side-by-side to collect integrated 2-BE samples during 7 trials for a total of 20 VOC and 14 2-BE samples.

Direct reading instruments (real-time air monitoring)

Real-time measurements of VOCs were collected with PIDs equipped with 10.6 eV lamps:

• ppbRAE 3000 (Model PGM-7340, Honeywell International Inc., Charlotte, North Carolina, USA). This instrument features a part per billion (ppb) sensor with a reported detection range of 0 to 9,999 ppb.

• RKI GX-6000 (RKI Instruments, Union City, California, USA). This instrument features both a ppb and a ppm level sensor, with detection ranges of 0 to 50,000 ppb and 0 to 6,000 ppm, respectively.

Calibration of the PID sensors was performed with 10 ppm and 100 ppm isobutylene. PIDs were configured to record VOC concentrations at 1-s logging intervals. The datalog recorded by ppbRAEs contained values that were out of the sensor range. These data points were removed before averaging the data into minute-long averages per guidance from Honeywell RAE Technical Support Associate (3 Aug 2021 email from L Santana).

The minute-long averages were then used to calculate task duration average VOC concentrations. Task duration refers to the estimated duration during which consumers are expected to be potentially exposed to vapors from using the product once, adjusted to active air sampling and analytical method requirements. The task duration for FCP spraying was estimated to be 3 min and was 15 min for GC spraying and wiping based on manufacturer’s instructions and online demonstrations of product use.

Active sampling (integrated sample collection)

Airborne samples of acetone and IPA were collected using 1.4-L minicans according to the EPA TO-15 sampling and analytical method (“Compendium of Methods” 1999) for the FCP spraying task. Samples were analyzed at an American Industrial Hygiene Association (AIHA) and National Environmental Laboratory Accreditation Program accredited laboratory. The results for acetone and IPA were background corrected before summarizing the data. In addition, other compounds reported in the EPA-TO 15 lab report were not included in the analysis as they were not listed as components of the product in the SDS.

Airborne samples of 2-BE were collected according to the National Institute for Occupational Safety and Health Manual of Analytical Methods 1403, Issue 3 (2003) using sorbent tubes containing coconut shell charcoal for the GC spraying and wiping task. Samples were collected with Gilair (Sensidyne LP, St. Petersburg, Florida, USA) pumps calibrated with a DryCal Defender 510 (Mesa Laboratories, Inc., Lakewood, Colorado, USA). Pumps collected air for 15 min at 0.2 LPM, the maximum recommended flow rate for 2-BE sampling. Samples were analyzed at an AIHA-accredited laboratory. Where results were below the limit of detection, the laboratory reporting limit of 2 ug was used as the measured concentration value and then averaged for each scenario.

Emission rate characterization

Total emitted mass was estimated by weighing the mass of the aerosol can prior to and upon completion of each trial using an Ohaus Adventurer Pro AV2102 Laboratory Balance (OHAUS Corporation, Parsippany, New Jersey, USA) for each application. Similarly, the weight of the surface which was intentionally sprayed as part of the task and the weight of the materials used for capturing overspray or used for wiping were also recorded before and after each trial to account for mass of product that was not airborne.

Experimental conditions

Air flow was measured at the face of each lab exhaust opening and at the face of the LEV opening using the VelociCalc Multi-Function Ventilation Meter 9565 (Model 9565-P, TSI, Shoreview, Minnesota, USA). Air exchange rate (AER) in the experimental area was then estimated by determining the volumetric flow rate of air out of the lab with the general ventilation system only and then again with the general ventilation system plus the LEV. Temperature and relative humidity (RH) were monitored throughout the trials using the VelociCalc with an adaptable TSI Indoor Air Quality 982 probe.

Literature search

A literature search was conducted to identify exposure studies (both worker and consumer) assessing the use of fabric protector and glass cleaning spray products. Studies that reported air concentrations for VOCs, acetone, IPA, and/or 2-BE during the use of these products were included for comparison to the data collected during the robotic spraying trials. Studies which reported only full shift or 8-h and longer measurement data were excluded from our analysis.

Data analysis

Air sampling results were summarized to characterize the range, average, and variability of the air concentrations for both tasks. Exploratory analysis was conducted to (i) compare the air concentration data measured in this study during robotic simulations and those reported in literature for humans using similar products and (ii) evaluate the impact of exposure modifying factors during spraying activities. No statistical tests were conducted to exam the differences between the results from this study and those reported in the literature.

RESULTS

Summary of measured air sampling data

For FCP spraying, acetone concentrations ranged from 0.012 to 4.5 ppm, with an average of 0.88 ppm; IPA concentrations ranged from non-detect to 1.4 ppm with an average of 0.26 ppm; and total VOCs measured with PIDs ranged from 0.41 to 4.5 ppm, with an average of 1.57 ppm. For GC spraying and wiping, 2-BE concentrations ranged from non-detect to 0.19 ppm with an average of 0.15 ppm and total VOCs measured with PIDs ranged from 0.007 to 0.34 ppm with an average of 0.17 ppm.

Air concentrations for all substances were lower during trials with LEV, with the exception of 2-BE for GC spraying and wiping (0.16 ppm with LEV, 0.15 ppm without LEV) (see Table 2). The difference was small and may be due to the fact that the amount of 2-BE detected in air was near the reporting limit, and in 3 samples, collected during trials 5, 6, and 7, less than the reporting limit. In addition, the LEV may have been positioned too far away from the spraying to effectively capture the vapors.

Table 2.

Air sampling data for Tasks 1 and 2 compared to DNELs and ACGIH TLVs.

	Substance	Average air concentration without LEV in ppm (SD)	Average air concentration with LEV in ppm (SD)	Long-termb inhalation General Population DNEL (ppm)	Acutec inhalation General Population DNEL (ppm)	ACGIH 8-h TLVd (ppm)	ACGIH 15-min STELe (ppm)
FCP spraying	Acetone	1.53 (1.57)	0.24 (0.26)	84.2	NA	500	750
	IPA	0.42 (0.48)	0.10 (0.09)	36.2	NA	200	400
	VOC (PID)	2.42 (1.49)	0.71 (0.26)	NA	NA	NA	NA
GC spraying and wiping	2-BE	0.15 (0.02)	0.16 (0.02)	12.2	88.1	20	NA
	VOC (PID)	0.19 (0.05)	0.14 (0.05)	NA	NA	NA	NA

^aAverage air concentrations in this table are task duration average concentrations (i.e. time-weighted concentration over 3 min for Task 1 and 15 min for Task 2). When comparing to long-term DNELs or 8-h TLVs, long-term average concentrations would be substantially lower, unless products were continuously used during the averaging time.

^bLong-term inhalation DNELs apply to repeated inhalation exposures.

^cAcute inhalation DNELs apply to occasional (minutes-hours) inhalation exposures.

^dACGIH 8-h TLVs are exposure limits that should not be exceeded during an 8-h workday.

^eACGIH 15-min STELs are exposure limits that should not be exceeded at any time during a workday over a 15-min period.

Emission rates

For FCP spraying, the total mass of product released from the can during each trial was on average 12.97 g (±0.55 g). Based on the total mass and the average % concentration of substances reported in the SDS, the estimated average mass of total VOCs in air was 7.00 g (±0.29 g), 5.06 g (±0.21 g) acetone, and 4.28 g (±0.18 g) IPA. For GC spraying and wiping, the total mass of product released during each trial was on average 4.14 g (±0.63 g). Based on the total mass and the average % concentration of substances reported in the SDS, the estimated average mass of total VOCs in air was 0.39 g (±0.06 g) and 0.26 g for 2-BE (±0.04 g). Additional data is available in Supplementary Table S1. Average substance emission rates were calculated using the estimated mass of substances in air and are summarized in Table 3.

Table 3.

Average emission rates.

Task	LEV	Average acetone emission rate in g/min (SD)	Average IPA emission rate in g/min (SD)	Average 2-BE emission rate in g/min (SD)a	Average VOC emission rate in g/min (SD)
FCP spraying	Yes	28.85 (1.26)	24.43 (1.05)	NA	39.94 (1.72)
	No	27.33 (0.49)	23.11 (0.40)	NA	37.85 (0.66)
GC spraying and wiping	Yes	NA	NA	2.90 (0.10)	4.24 (0.24)
	No	NA	NA	2.60 (0.18)	3.64 (0.74)

^aFor comparison to air sampling data, the 2-BE emission rate is based on product mass released in air during trials 1–7 (2-BE was only sampled for during trials 1–7 whereas VOCs were measured with PIDs for all 10 trials).

Experimental conditions

Average temperature and RH during the trials was 73.5 °F (72.7–74.6 °F) and 47.9% RH (38.1–51.1% RH). The volumetric flowrate of air out of the experimental area (removal rate) with general ventilation was approximately 20.6 m³/min. The volumetric flow rate of the LEV was estimated to be 1.6 m³/min. Using this data, the air removal rate in the experimental area with LEV was calculated to be 22.2 m³/min. Using an experimental area volume of 26 m³, the AER in the experimental area was estimated to be 6.8 h⁻¹ when LEV was used and 3.2 h⁻¹ when LEV was not used.

Data reported in the literature

Four studies (Vincent et al. 1993; Singer et al. 2006; Bello et al. 2010, 2013) where cleaning products containing 2-BE were used similarly to the robotic GC spraying and wiping task were identified. Where the studies reported air concentration data for multiple cleaning tasks, results for the tasks that most resembled the robotic simulation were included for comparison as described below.

In Vincent et al. (1993), 4 worker groups performing cleaning tasks were evaluated. The cleaning tasks performed by the Office Cleaners was considered to be most similar to the glass cleaning task simulated by the robotic arm. Air sampling results for Office Cleaners who used formula A (product containing 9.8% 2-BE) showed an average concentration of 0.32 ppm (range <0.3–0.73 ppm). Air sampling results for Office Cleaners who used formula B (product containing 0.9% 2-BE) were <0.30 ppm (all samples less than the limit of detection).

In Singer et al. (2006), 5 different tasks were simulated with a variety of cleaning products. The “Counter cleaning, spray and wipe only” task was most similar to the task performed by the robot in our study. Average 2-BE concentrations were 330 ug/m³ (0.068 ppm) for the GC (product containing 0.6% 2-BE) and 1410 ug/m³ (0.292 ppm) for the general-purpose cleaner (product containing 2.6% 2-BE).

In Bello et al. (2010), 3 cleaning tasks were simulated using a variety of cleaning products. Mirror cleaning was considered to be most similar to the robotic spraying and wiping task. Average 2-BE air concentrations were 2.96 ppm to 13.08 ppm (small bathroom cleaning without ventilation; product containing 1–2% 2-BE and 6–10% 2-BE, respectively) and 0.32 to 1.98 ppm (large bathroom cleaning with ventilation; product containing 1–2% 2-BE and 6–10% 2-BE). Total VOC concentrations ranged from 0.02 to 5.26 ppm.

In Bello et al. (2013), 3 cleaning tasks were simulated using a variety of cleaning products. Mirror cleaning was considered to be similar to the robotic spraying and wiping task. 2-BE air concentrations ranged from 0.1 ppm to 8.7 ppm for a variety of scenarios (small and large bathroom, with and without ventilation, varying concentrations of 2-BE in product). The authors noted that average total VOC concentrations ranged from 88 to 2910 ppb; however, task and product-specific total VOC concentrations were not reported.

DISCUSSION

Exposure measurement summary

In this study, air monitoring data were collected for estimation of personal inhalation exposures during the use of 2 spraying products. To ensure protection of consumer and worker health, exposures are often assessed against benchmark levels or exposure limits, such as derived no effect levels (DNEL) and American Conference of Governmental Industrial Hygienists (ACGIH) Threshold Limit Values (TLV). DNELs are levels of exposure above which humans (general population or worker) should not be exposed and are used by manufacturers or importers of chemicals to identify conditions under which products can be safely used ([ECHA] European Chemicals Agency 2012). ACGIH TLVs are airborne concentrations of chemical substances under which it is believed that nearly all workers may be repeatedly exposed, day after day over a working lifetime without adverse health effects (TLVs and BEIs 2022). Results from this study show that inhalation exposures to acetone and IPA during FCP spraying and to 2-BE during GC spraying and wiping are low when compared to general population DNELs and ACGIH TLVs.

Side-by-side sample comparison

The results of the side-by-side integrated air sampling data varied greatly for FCP trials but not for GC trials. The variability observed for the FCP spraying task may be attributed to the short sampling duration of 3 min and also the volatility of the substances measured (acetone vapor pressure is 30,798 Pa and IPA vapor pressure is 6,058 Pa). The higher vapor pressure would have resulted in the VOCs being released more quickly and differences in local dispersion of the spray cloud. In addition, the air inlet tube on the minican samplers were further away from the spraying than the PIDs (see Fig. 1a). In comparison, the GC task sampling duration was 15 min, the vapor pressure of 2-BE is 117 Pa, and the sampling equipment pulled in air from a similar distance from the spraying (see Fig. 1b). In future studies, modifications to the sampling strategy may help to address the large variability observed in side-by-side samples.

Evaluation of data found in literature

The average air concentrations for both 2-BE and VOCs collected during the robotic simulations for glass cleaning are within the range of data reported on similar cleaning tasks in the literature, but on the lower end (see Fig. 3).

Fig. 3.

Air concentration data for cleaning tasks reported in the literature and measured during robotic simulation.

Differences in exposure scenarios (e.g. room size, ventilation rate, %2-BE in product, product spraying, and use duration) as well as the sample collection strategy, including the sampling and analytical method, sample duration, and sampling equipment location may explain why results from the robotic simulations were on the lower end of results reported in the literature. For example, the highest reported 2-BE concentration was 13.08 ppm by Bello et al (2010). In this scenario, mirror cleaning was done inside a small (4.56 m³) unventilated bathroom using a product containing 6–10% 2-BE; in the robotic simulation, the experimental area was approximately 26 m³ with a minimum AER of 3.2 ACH and a product containing 2–10% 2-BE. It is likely that both ventilation and product content contributed to the difference in air concentrations. As another example, during the robotic simulations, the air sampling equipment was located approximately 22 inches (about an arm’s length away) from the can, which resembles a user spraying the product as far as possible from his or her breathing zone (holding arm out straight, away from body). In reality, users (and those in the literature studies) may have held the product closer to their bodies, which would result in higher air concentrations in the personal breathing zone. Future robotic studies could evaluate the impact of this modifying factor by varying sample collection distances. Lastly, exposure duration is an important aspect of exposure and may have contributed to the differences in the air concentrations; this includes how one defines exposure duration. In this study, the air sampling duration was equal to the exposure duration and represented potential exposure from both the active use of the cleaner as well as some secondary exposure. The total spraying time, where product was released from the can, was 6 s and the total glass wiping time was approximately 2 min. In the literature, spraying time was not reported but product use time which is expected to include both spraying and wiping ranged from 2 min (Singer et al. 2006) to 15 min (Vincent et al. 1997); sampling duration ranged from instantaneous to 5 h. In one instance, product use duration was not reported (Bello et al. 2010) and in another, the product use duration was reported as 15 min with a sampling duration of 2.5–5 h (Vincent et al., 1993). Where the reported use time was significantly less than the sampling duration, air sampling concentrations would be expected to be less, but this would depend on other variables including how much product was released and the sample location.

A comprehensive table with 2-BE and VOC air sampling data from the studies listed above along with the data collected in this study is available in Supplementary Table S2. No studies with measured air concentration data for acetone, IPA, or VOCs during the use of fabric protector spray were identified.

Robotics for spraying and wiping

In this study, we successfully programmed a robotic arm to mimic human use of 2 consumer aerosol can products. The robotic spraying and wiping motions resembled the motions that humans would make to use similar products. More details on the hardware and software (coding) used for this study, as well as a discussion on strengths, limitations, and other considerations for selecting a robot to generate realistic human motions for use in exposure studies can be found in Supplementary Informations 3 and 4).

Strengths and limitations

In this study, we demonstrate that a robot can be used for generating exposure data both safely and repeatably. No human subjects were needed for this study, eliminating the concern of intentional exposures. In terms of repeatability, each trial consisted of the same amount of spraying and motion. The robot held the spray can trigger down for the desired amount of time and sprayed the intended surface area from the same distance, with the same speed, and with the same pattern for each trial. These controlled, repeated motions can be used to systematically evaluate variability in exposures caused by exposure-modifying factors such as room volume, AER, and product compositional differences, as well as human behavioral differences through simple experimental design and/or programming modifications. For example, humans may use different amounts of product to complete the same task; in future studies, trials could be repeated where all conditions are kept the same except for how long the robot sprays product.

Though the robot was able to hold the trigger down for the same amount of time during each trial, there was a slight decrease in the amount of product released from trial to trial. This is likely due to emptying of the can and therefore less pressure use after use. In a future study, if the amount of product released per use is a factor to be evaluated, additional cans of product with similar starting mass could be used to have multiple trials where the same amount of product is released.

Other limitations of this study are that trials were not performed in a controlled environment and the AERs inside the experiment area (see Fig. 2) were estimated using air flow out of the entire UT Austin HCRL based on measurements taken at all air exhaust openings including the LEV. The use of the box to capture overspray may have impacted the dispersion of the substances released from the can and the location of the LEV may have been too far away to be effective. In future studies, AER could be determined more accurately using tracer gas or decay curves using small particles and the capture efficiency of the LEV should be verified. Though the temperature and RH were monitored and relatively stable during the trials, the lab was notably colder and less humid on 1 experimental day. All of this may have contributed to the variability observed in the side-by-side samples for FCP spraying and the comparability to the literature studies on GC spraying and wiping. Though there is variability expected in real life, future studies could be conducted on a flat surface in an environmental chamber to ensure more realistic spatial dispersion of the airborne vapor and to control temperature, RH, and air flow. This will improve the understanding of impact from these variables. Last but not least, for the GC task, the average amount of 2-BE detected in air was low (near the reporting limit and in 3 samples, less than the reporting limit) and using the laboratory reporting limit as the detected value will have biased the results high. This, in addition to the positioning of the LEV, made it difficult to assess the impact of LEV for this task.

Fig. 2.

University of Texas at Austin, Human Centered Robotics Lab (not-to-scale).

CONCLUSIONS

We successfully programmed a robot to mimic human spraying and wiping tasks and generated simulated breathing zone measurements within exposure ranges reported in the literature. In addition, the programming and hardware developed and selected in this study expands the options for collecting exposure data for spraying activities and can serve as a platform to be reused by other researchers for generating other activity-specific exposure data, opening the door to a significant increase in consumer and occupational exposure information. This study highlights that there is a certain amount of unexplained variability in exposure measurement data. Robots can be used to test various hypotheses and understand what is driving this variability in an efficient, systematic way. With rapidly evolving robotic technologies, more complex tasks and variability driven by use conditions as well as user behavior will become easier to assess. Mobile robots that can be teleoperated or easily programmed along with more sophisticated vision and force sensors will allow for evaluation of a broad range of user behaviors and conditions including improper and proper use of products, spraying on multiple surfaces, use of different spray equipment (aerosol can, trigger spray, and spray gun), and use of products in various settings such as bathrooms, kitchens, or even commercial buildings and spaces.

In a future publication, we will compare the measured data from this study with model predictions to help corroborate or improve existing exposure models to gain confidence in their application to support product registration and other safety and health risk assessments.

Supplementary material

Supplementary material are available at Annals of Work Exposures and Health online.

1. Table S1. Estimated mass of substances in air for each trial.

2. Table S2. Summary of 2-BE and VOC air concentrations reported in literature during glass cleaning.

3. Supplemental Information 3. Details on hardware and software development.

4. Supplemental Information 4. Considerations for using and selecting a robot for use in exposure studies.

5. FCP Acetone and IPA results (https://osf.io/fqhbe/).

6. GC 2-BE results (https://osf.io/fqhbe/).

7. FCP VOC data (measured with PIDs) (https://osf.io/fqhbe/).

8. GC VOC data (measured with PIDs) (https://osf.io/fqhbe/).

9. Videos of the robotic simulations are available online (https://youtu.be/IGOh9xwLeI4; https://youtu.be/B192m3QB4h0; and https://youtu.be/DYj4F5bv5nE).

Author contributions

JS designed and directed the project, collected exposure data, and prepared the manuscript; HQ and SIM provided input to exposure data analysis and discussion. JEL designed and created the robotic end effector attachment, programmed and operated the robot during trials, and provided robotic hardware and software design details for the manuscript; LS supervised the design, programming, and operation of the robot; SIM and LS provided technical guidance and input to the manuscript.

Funding

The research was funded by ExxonMobil Biomedical Sciences, Inc. (EMBSI). The Apptronik Scorpio arm was acquired through an Office of Naval Research Defense University Research Instrumentation Program (ONR DURIP) grant.

Conflict of interest statement.

Three of the authors of this paper are employed by ExxonMobil Biomedical Sciences, Inc. (EMBSI) who provided funding for this study. In addition, the Apptronik Scorpio arm was acquired through the Office of Naval Research Defense University Research Instrumentation Program (ONR DURIP) Grant #N00014-18-1-2238.The authors designed and executed the study and have sole responsibility for the writing and content of the manuscript..

Data availability

The exposure data underlying this article are available in Center for Open Science at https://osf.io/fqhbe/ and in its online supplementary material, tables S1 and S2. The robotic data underlying this article are available in its online supplementary material, Supplemental Information 3 and Supplemental Information 4; additional hardware and software development details will be shared on reasonable request to the corresponding author.