|
Risk Assessment Methods for Autonomous Agricultural Machines: A Review of Current Practices and Future Needs
John M. Shutske1,*, Kelly J. Sandner1, Zachary Jamieson1
Published in Applied Engineering in Agriculture 39(1): 109-120 (doi: 10.13031/aea.15281). Copyright 2023 American Society of Agricultural and Biological Engineers.
1Biological Systems Engineering, University of Wisconsin, Madison, Wisconsin, USA.
* Correspondence: shutske@wisc.edu
The authors have paid for open access for this article. This work is licensed under a Creative Commons Attribution-NonCommercial-NoDerivatives 4.0 International License https://creative commons.org/licenses/by-nc-nd/4.0/
Submitted for review on 27 July 2022 as manuscript number MS 15281; approved for publication as a Research Article by Associate Editor Dr. Michael Sama and Community Editor Dr. Heping Zhu of the Machinery Systems Community of ASABE on 9 January 2023.
Highlights
Abstract. Technology continues to advance in agricultural machines and includes the development of highly automated, robotic, autonomous, and other types of machines used in fields, farmsteads, buildings, and other farm production locations. New engineering design and safety-related standards have been developed in the past half-decade, but safety remains a concern of key stakeholders and is a barrier that could influence widespread adoption. A survey of practicing engineers and researchers involved with highly automated and autonomous agricultural machine design will be presented that shows the methods for risk assessment and control currently in use including different frameworks for hazard and failure identification, prediction, and quantification. The use of engineering design standards (ASABE, ISO, and others) among practitioners is discussed including some important needs that go beyond obstacle detection and injury prevention for operators. These include safety and risk issues connected to animals, property, civic infrastructure, downtime, cyber, and environmental risk. Commonly used risk assessment methods such as the related failure modes and effects analysis (FMEA) or hazard analysis and risk assessment (HARA) are a useful starting point but are based on historical data and experience that can be used to estimate the probability and severity levels of undesirable failures or incidents such as injuries. These data do not yet exist as compared to risk assessment data that can be used to assess incident occurrence probability, failure, detectability, or controllability in more traditional machines. Suggestions are presented for further development of standards and practice recommendations including software needs and operational data that might be used by autonomous machines that is informed by what we do know about past farm incidents that could include accidents, injuries, and other unexpected failures.
There is a clear trend toward increased levels of automation and autonomy in production agriculture. Xu (2020) explains the difference between the terms automated versus autonomous by saying, “Automation is the ability of a system to perform well-defined tasks and to produce deterministic results, relying on a fixed set of rules and algorithms without AI technologies.” The author further explains, “autonomy specifically refers to the ability of an AI-based autonomous system to perform specific tasks.” Autonomy implies that operations are carried out without the need for a human operator to be immediately present. The forces that contribute to the trends toward both automation and autonomy include issues connected to labor cost and availability (Castillo, 2020). An additional fact is that many tasks that are common in agriculture are associated with high levels of drudgery (Dash et al., 2021; Kootstra et al., 2021; Rakhra et al., 2022). This leads to a logical conclusion that many agricultural work tasks that currently require a human “operator” might benefit from further levels of automation (up to and including fully autonomous operation) if the tasks and conditions warrant. Other factors that appear to be technology drivers include the need to address global food security; the potential for automated technology to help in mitigating climate change; and the role of new sensor-based technologies in environmental protection through a reduction of input use such as pesticides and nutrients (Lowenberg-DeBoer et al., 2022).
Reducing Injury Risk by Reducing Operator Exposure
Production agriculture is a dangerous occupation considering high rates of fatal and non-fatal injuries (National Safety Council, 2022). Machinery and tractors are among the hazards often involved in fatal injury events based on a recent analysis of detailed incident data collected in key agricultural states (Li et al., 2022). One potential benefit of removing the human operator from a mechanized system is to reduce human exposure and associated risk that contributes to high rates of injury associated with farm machinery operation. The most preferred way to reduce injury risk is to design machine systems that reduce or eliminate direct human operator hazard exposure (Wogalter, 2018). As many farm work-related fatalities involve tractors, “autonomous” tractors and other highly automated machine forms have the potential to reduce injuries to humans from events such as tractor overturns, machine runovers, entanglements, and other common forms of injury or death.
Other Types of Risk/Exposure
While there is a strong potential to reduce operator risk through automation and autonomy, this does not mean safety issues will be eliminated. Gerberich et al. (1998) showed that with traditional farm machines, injuries often occur during repair, adjustment, and maintenance tasks. A machine that operates in an autonomous mode or with various levels of operator involvement will present different types of risks and scenarios. Risk will not simply disappear. In addition to operator and repair/maintenance personnel risks, we also must consider exposures to bystanders in a field, building, farmyard, or other area in which an autonomous or highly automated machine is operating. There is also a potential risk to the “public” if a machine were to perform in an expected manner due to a software or sensor failure causing a situation like leaving field boundaries and entering other areas or physical property including public roadways.
Human risk from traumatic injury is not the only safety risk. A machine that autonomously performs an application procedure with fertilizer, livestock manure, pesticides, or other product presents a potential environmental risk if an error or malfunction were to occur and then lead to a spill, overturn, or application in the wrong area or in certain conditions (like herbicide application in undetected windy conditions). A failure in the field or other operating location such as an overturn would likely lead to a costly downtime incident and/or damage to the machine or other property, even if a human operator is not injured. Costs would be incurred as a result of operational delay while conditions are corrected, and machines get repaired. Additionally, autonomous machines must not only sense people (operators, repair personnel, bystanders), but they must also consider potential exposure and risk to animals such as livestock, wildlife, and pets. Those who design, purchase, use, and service these machines must also consider issues of cyber- and software security (Freyhof et al., 2022). A machine that is tampered with or “hacked” could perform in unexpected and unsafe ways leading to multiple human and economic risks. The downtime caused by an intentional software breach on a farm machine could have catastrophic economic impact for a farmer or agricultural service provider, especially with operations that are time sensitive.
Little is published on these human and other risks (and benefits) associated with agricultural machine autonomy. At least one early reference seemed to downplay safety-related concerns. The textbook chapter in Agricultural Automation: Fundamentals and Practices (Zhang and Pierce, 2013) by chapter author Noguchi on “vehicle robots” opined, “If the robot must take full responsibility in the unlikely event of an accident, it will lead to enormous cost increase and hinder the progress of robotization.” This statement both surmises that probability of a safety-related issue is low in the authors’ opinion, while also stating that if something were to occur, consequences would be significant. In a recently released “Code of Practice” published by Grain Producers Australia (GPA), Tractor and Machinery Association (TMA), and the Society of Precision Agriculture Australia (SPAA) Rainbow (2021) writes, “Farming with mobile machinery that has autonomous functions, like any agricultural activity, is hazardous with many inherent risks. When integrated with a manned farm operation, additional risks may be present beyond those recognized for conventional farming techniques.”
Key Strategies to Evaluate and Control Risk
There are multiple methods used to address these types of cited safety concerns. Often, this collection of methods is referred to as risk “assessment,” an evaluative process that includes identifying hazards, predicting ways in which a component or system could fail or be misused, evaluating the outcomes if undesired events occur, and examination of other variables. Bahr (2015) explains that risk assessment processes help engineers and others to prioritize control measures. Once identified and prioritized, safety engineers must consider a range of risk hazard control options designed to mitigate risk.
There are multiple ways to mitigate safety-related hazards, conditions, and associated risk. Several sources cite a hierarchy of control strategies (Barnett, 2020). Eliminating hazards through design is most preferred within this hierarchy. Hazard elimination is followed by or done in conjunction with the design and application of “safeguard” devices. An example with traditional machines is risk reduction associated with tractor overturn deaths. Design features that eliminate much of the hazard can include a machine with a very wide baseline of stability and an extremely low center of gravity that would lead to it being nearly impossible to roll a tractor in normal farm conditions. This is an example of “designing out” the hazard. Yet, since designing out all hazards is not always fully practical or can be overly costly and impractical, safeguarding devices include things like rollover protective structures and seatbelts.
The work reported here concentrates on evaluating risk and controlling hazards as early as possible in the design of machines. With modern designs, engineered safeguard devices are also appropriate (guards, interlocks, sensors, control systems). Other hazard and risk mitigation measures involving the operator are not considered in this research. These would include education (to improve operator human knowledge), as well as safety warnings, instructions, and other information. These approaches that rely on conveying information to the operator to ensure safety remain necessary but are often not an adequate or fully effective means of safety engineering by themselves.
A key to the safe design of any machine is engineering standards promulgated by groups like ASABE, ISO, ANSI, and other standards-making organizations. Sometimes, standards incorporate, recommend, or suggest systematic risk assessment methods. For those engaged in the design process, several risk assessment techniques are commonly used or referred to within standards. A more detailed explanation of risk assessment methods is contained in the textbook by Bahr (2015). A sampling of these includes:
- Hazard Analysis
- Failure Modes and Effects Analysis (FMEA)
- Fault Tree Analysis (FTA)
- Hazard Analysis and Risk Assessment (HARA)
- Hazard and Operability Analysis (HAZOP)
In the work described in this article, we:
- Identify and evaluate existing engineering design and safety standards with the goal of identifying specific risk assessment methods incorporated within these standards while also identifying the possible gaps in the risk assessment coverage provided by standards.
- Describe a survey of industry and other engineering professionals engaged in the design processes for highly automated or autonomous agricultural machines. This survey queried participants about the methods they use for risk assessment and control including methods explicitly addressed in standards or existing outside of standards such as FMEA, FTA, and others.
- Discuss gaps, needs, and opportunities to improve design-related tools such as standards and risk assessment and control methods to aid engineers and others in the design process so that safety concerns are lessened and hopefully will present less of a barrier to adoption for the industry and to the stakeholders who might benefit from new advances in automated technology.
Methods
Standards Analysis
The first step in identifying and analyzing safety and design standards was to identify as many existing standards connected to the safe design of highly automated or autonomous agricultural machines as possible. This was and remains a challenge because of the rapidly changing nature of the industry, with many standards being discussed, developed, and in various stages of the promulgation process at any given point in time. As an example, the standard ISO 18497 (Agricultural machinery and tractors—Safety of highly automated agricultural machines—Principles for design) was originally released for use in 2018 but is now in the early phases of a major revision and this revised version will not be released for use until 2023 or 2024. In identifying hazards and conducting other parts of this study, the official “definition” of automation and autonomy was challenging as there is not yet a uniform set of definitions available at this time that were located by the authors. ISO 18497 (2018) used the term “Highly Automated Agricultural Machine” or HAAM, defined within the standard as, “autonomous mobile vehicle or machine with or without on-board operator allowing highly automated operation.” We also acknowledge that new definitions and frameworks for automated and autonomous machines and functions are forthcoming through future ISO standards that are in the early stages of revision or development. For the sake of the work reported in this article, authors relied on industry-based frameworks that included the “Automation Defined by Case IH” categorization shown in figure 1.
 |
| Figure 1. Levels of automation as defined by CASE IH (2018). |
Additionally, we considered the contributions of the automotive industry, through the Society of Automotive Engineers standard SAE J3016(SAE, 2021). This standard categorizes six levels of automation used to describe characteristics of on-road autonomous vehicles based on specific features and decreasing requirements of human interaction. While this standard is intended for road-operated motor vehicles, for the sake of this research, the authors found these levels to be useful to help evaluate and consider the range of possibilities in agricultural machine automation/autonomy from “none” to “full.” We acknowledge that this framework may be less than optimal for agricultural machines in the long term especially as ISO 18497 evolves.
To identify available standards, a search was conducted through the University of Wisconsin-Madison library system and its specialty databases for engineering standard searches. These included:
- ASABE Technical Library – Includes all ASABE standards
- SAE Mobilus – SAE Standards
- Techstreet Enterprise – Includes ASME, ISO, ANSI, ASCE
We also identified and reviewed an important document by Rainbow (2021) developed in Australia. This document contains a detailed list of engineering standards pertinent to the safe design of highly automated agricultural machines in its appendix. In addition, informal advisory conversations were done with a standards experts from ASABE, private companies, and a representative from machinery company trade organization to make our search as complete as possible. To protect privacy of individuals and since these conversations were conducted solely to make sure we had located as many standards as possible, the conversations are not cited. They occurred in 2020 and 2021.
After identifying and locating copies of standards, an independent analysis was conducted with each standard by three individuals. They included: a graduate student studying risk in automated agricultural machinery, an agricultural safety specialist and engineering faculty member, and an undergraduate student. The analysis included examination of characteristics including the standard’s date, scope (areas in and out of coverage), risk analysis methods described in the standard, other significant information that might lead to better understanding of the standard’s coverage.
Survey of Engineering Design Personnel
The purpose of this survey was to assess where industry and other engineers are relative to their use of various risk assessment methods. This included both the use of formal methods (FMEA, HARA, FTA, etc.) as well techniques that are referenced within a subset of standards that include some coverage of risk assessment methods or references to formal techniques.
The survey was developed using the Qualtrics XM software package (Qualtrics, Provo, Utah) that allows users to develop questions, program logic flows, track distribution, and analyze collected data for desktop computer or mobile response. Survey recipients were presented with a series of definitions of relevant terminology that would be used in the questions including “autonomous,” “remote operator,” and “UAS” (UAS, or unmanned aircraft systems, was identified only to let participants know that we were excluding UAS in this study). Recipients were asked to consider the questions within the scope of the agricultural industry rather than other industries that larger companies might serve (such as mining, construction, or forestry). Additionally, to help define concepts of automation, recipients were presented with the graphical image depicting the levels of automation as defined by the previously described CASE IH efforts as was shown in figure 1. This definition was used because it was in an easy-to-understand format that included both graphic and plain text descriptions. For this survey, recipients were directed to only consider machines which fell under levels 4 and 5 of the figure 1 classification system; cases in which there were no humans directly interacting with or controlling machines.
The survey was distributed to selected individuals and consisted of 29 possible questions, 11 of which required responses and 18 optional questions which provided respondents the opportunity to further elaborate on their answers to the required questions. Input validation within Qualtrics was used to verify that a response was given for the 11 core questions for the respondent to progress within the survey. A copy of the survey instrument is provided in the supplemental materials. The first question of the survey asked recipients if they have ever or currently worked with the design, development, testing, application, research, or development of autonomous agricultural machinery. Simple Boolean logic was used to filter the respondents that indicated they did not meet these criteria, directing them to the end of the survey.
This filtering process was used to ensure that only individuals who have worked with highly automated agricultural machinery would provide data. This allowed the data collected to be an accurate representation of variations in the processes used to conduct risk assessments of traditional, non-autonomous, machinery, and novel highly automated machinery. The survey also explored the application of engineering design and safety standards that were deemed by the research team to be most used by engineers while working with highly automated machinery based on standard scope and detailed coverage of human and other forms of risk.
The initial list of survey recipients was identified by accessing the American Society of Agricultural and Biological Engineers (ASABE) committees list and identifying committees deemed by the research team to include a scope and activities most relevant to highly automated agricultural machinery. A list of the selected committees identified is provided in table 1. The team erred on the side of “over-selecting” recipients as we did not want to miss or exclude key respondents. However, a significant limitation of this methodology is that industry personnel who are not involved in ASABE or its committee activities were not included.
The name, title, organization, and email address of the members of relevant committees were exported to a Microsoft Excel file. Duplicate entries from individuals participating in multiple committees, were eliminated. The PI of this research project and two faculty members of the graduate student’s examining committee were also excluded. Techniques recommended by Dillman (2011) were used. The identified individuals were emailed a total of three times with a link to the survey and additional information. The initial request included a cover letter and IRB-required consent information. Two reminder emails were sent; one after the first full week, and one three weeks post-initial contact, respectively. The initial and follow-up emails can be seen in the supplemental materials.
| Table 1. ASABE committees included in survey distribution. | |||
| ASABE Committee | Committee Name | ||
| ESH-01 | Ergonomics, Safety, and Health Executive/Oversight | ||
| ESH-03/1 | External Standards Development | ||
| ESH-04/2 | Farmers with Disabilities Technology Exchange | ||
| ITSC-254 | Emerging Information Systems | ||
| ITSC-312 | Machine Vision | ||
| ITSC-318 | Mechatronics and Robotics | ||
| MS-01 | Machinery Systems Executive | ||
| MS-03 | Machinery Systems Standards Oversight | ||
| MS-23 | Tractors and Machinery for Agriculture and Forestry/US TAG ISO/TC 23 | ||
| MS-23/14 | Ag Machinery – Symbols, Displays, Manuals/US TAG ISO/TC 23/SC14 | ||
| MS-23/19 | Ag Electronics/US TAG ISO/TC 23/SC 19 | ||
| MS-23/19/1 | Applications and Data Interfaces | ||
| MS-23/19/5 | Communication Infrastructures | ||
| MS-23/2 | Ag Machinery - Common Tests/US TAG ISO/TC 23/SC 2 | ||
| MS-23/2/1 | Environment Within Ag Vehicle Enclosures | ||
| MS- 23/3 | Ag Machinery - Safety and Comfort/US TAG ISO/TC 23/SC3 | ||
| MS-23/4 | Tractors and US TAG ISO/TC 23/SC 4 | ||
| MS-23/4/5 | Tractor Implement Interface/PTO | ||
| MS-23/7/2 | Forage & Biomass Engineering | ||
| MS-23/7/3 | Cotton Engineering | ||
| MS-47 | Distinguished Lecture Series - design of components for agricultural tractors and self-propelled machines | ||
| MS-49 | Crop Production Systems, Machinery, and Logistics | ||
| MS-54 | Precision Agriculture | ||
| MS-58 | Agricultural Equipment Automation | ||
| MS-60 | Unmanned Aerial Systems | ||
| P-127 | Robotics Student Design Competition | ||
| AFS-403/1 | Milk Handling Equipment | ||
| PRS-702 | Crop & Feed Processing & Storage | ||
| X497 | Agricultural Machinery Management Data | ||
As was previously mentioned, the study recruitment protocol and survey tool were presented to the Minimal Risk Research Institutional Review Board (MRR IRB) in accordance with University and Federal policy. The study passed initial review and was granted exemption from further review. Participating in this study presented a possible risk to respondents if a company was sensitive about disclosing the techniques and methods used to assess risk in relation to highly automated agricultural machinery. To mitigate this potential risk, the aggregated list of potential survey respondents was stored separately from other study data under two-factor authentication required for accessing the data. Additionally, participants were emailed an anonymous link to access the Qualtrics survey that prevented any ability to connect individuals to their survey responses. Participants were informed about all aspects of the study during the consent process and had the option to withdraw from any activity that made them uncomfortable at any point. Respondents who completed the survey were required to email the study team separately to receive an incentive/compensation for their time. Compensation was a fact sheet that summarized the results of the survey as well as a compendium of existing highly automated agricultural machinery that had been compiled in the early phases of this project. The information of those that directly contacted the study team were stored under two-factor authentication, maintaining a separation from any other data, until the compensation was distributed, then that information was destroyed.
Statistically, the variation between techniques used to conduct risk assessment for non-autonomous machinery and highly automated machinery was of key interest to the study team. To quantify differences, a frequency analysis was performed using Microsoft Excel. The data were exported to Microsoft Excel separate the individual data points for analysis. Respondents were allowed to choose more than one option for some of the questions, the data analytics provided by Qualtrics resulting in a varying denominator that needs to be considered during analysis. Data were also analyzed statistically using SAS 9.4 (SAS Institute, Cary, N.C.). A frequency analysis of the variables was performed to check for trends.
Results
Standards Analysis
Table 2 shows a summary of the 13 standards that were examined and summarized. The first six standards are perhaps the most applicable in addressing the safety and risk-related issues of interest within this study. This includes the issues of human injury to operators, bystanders, those involved in repairs and maintenance, and the general public. This project also includes examining other risk issues as were defined in the introduction section (risk to animals, the environment, downtime, cyber, and others). However, little information is found in existing standards beyond safety issues and hazards for operators and others in field conditions.
Here are key observations on a subset of the standards within table 2 based on work adapted from Sandner (2021).
ISO 18497: Agricultural Machinery and Tractors–Safety of Highly Automated Agricultural Machines– Principles for Design
The purpose of ISO 18497 is to provide specific, “…principles for the design of highly automated aspects of highly automated machines and vehicles (e.g. agricultural tractors, tractor implement systems, implements, and self-propelled machinery) during agricultural field operations.” As it relates to improving overall machinery safety, this standard is tightly constrained to “field operations” and does not apply to situations or circumstances in which human operators, service personnel, or others are injured on tractors and other machines such as while performing repair tasks, highway travel, or mounting/dismounting equipment. As previously mentioned, injuries during these operations have been well-documented and proven to be common. ISO 18497 explicitly excludes farmyard or barn operations, as well as operations on public roads. Obviously, highly automated agricultural machines (HAAM) will highly likely operate in these excluded zones, thus when analyzing overall farm safety with these machines, these gaps are important in this standard. Additionally, the standard does not speak to issues of safety associated with maintenance procedures, whether they are performed in the field or in a formal maintenance shop.
While operating in the field, a general requirement of operational procedures states, “It shall not be possible to enable highly automated operation without the perception system confirming that the hazard zone is obstacle-free.” A perception system is elaborated as referring to a system that is capable of detecting, “objects in the path of the HAAM, persons approaching the HAAM and the position of the HAAM relative to detected obstacles and the boundary of the working area.” The potential risk and failure modes of perception systems on highly automated agricultural machines in the presence of obstacles are also presented. It is important to note that with any machine performing field operations, “obstacles” are not the only hazard present. Sloping terrain is mentioned as a possible visual obstruction for a physical obstacle but sloping terrain can also lead to vehicle rollovers in certain conditions. This scenario is not addressed, and no guidance is given provided as it pertains to designing HAAM for safe operations on a slope.
| Table 2. Summary of standards analysis. | |||||
| Standard No. | Title | Date Published | Risk Analysis Method Cited* | Protected Entities Associated with Hazard Exposure | Specifically Excluded Entities with Potential Hazard Exposure |
| ISO 12100 | Safety of machinery– General principles for design– Risk assessment and risk reduction | 2010 | FMEA | “Annex B gives…examples of hazards, hazardous situations and hazardous events…to clarify these concepts and assist the designer in the process of hazard identification.” | Scope says: “It does not deal with risk and/or damage to domestic animals, property or the environment.” |
| ISO/FDIS 18497 | Agricultural machinery and tractors–Safety of highly automated agricultural machines–Principles for design | 2018 | HARA | Only applies to field operations. Addresses human hazards to operator and others – crush, shear, entanglement, electrical, ergonomic. Addresses issues associated with loss of control, or harm to “third persons.” | Excludes farmyard operations and highway travel. |
| ISO 25119-1 | Tractors and machinery for agriculture and forestry–Safety related parts of control systems–Part 1: General principles for design and development | 2018 | HARA and refers to ISO 12100 | Points back to scope and coverage in ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. |
| ISO 25119-2 | Tractors and machinery for agriculture and forestry–Safety related parts of control systems–Part 2: Concept phase | 2018 | HARA | Only applies to harm of people. | Excludes property. |
| ISO 25119-3 | Tractors and machinery for agriculture and forestry–Safety related parts of control system–Part 3: Series development, hardware and software | 2018 | Refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ISO 25119-4 | Tractors and machinery for agriculture and forestry–Safety related parts of control systems–Part 4: Production, operation, modification and supporting processes | 2018 | Refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ISO 4254-1 | Agricultural machinery–Safety– Part 1: General requirements AMENDMENT 1 | 2021 | HARA and refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ISO 13849-1 | Safety of machinery–Safety-related parts of control systems–Part 1: General principles for design | 2015 | HARA and refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ISO 13849-2 | Safety of machinery–Safety-related parts of control systems–Part 2: Validation | 2012 | FMEA, FTA. refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ANSI/ASAE S318.18 | Safety for agricultural field equipment | 2017 | Refers to ISO 12100 | In referencing ISO 12100 - risk and/or damage to domestic animals, property or the environment is not in scope. | |
| ISO 26322-1 | Tractors for agriculture and forestry–Safety–Part 1: Standard tractors | 2008 | HARA | ||
| ISO/TR 14121-2 | Safety aspects–Guidelines for their inclusion in standards | 2012 | HARA | ||
| ISO 10975 | Tractors and machinery for agriculture–Auto-guidance systems for operator-controlled tractors and self-propelled machines–Safety requirements | 2009 | None | ||
ISO 18497 does not deeply explore aspects of machine learning or artificial intelligence (AI). However it does state that a possible failure mode of the perception system may be due to, “inadequate experience, training, or validation of the classifier.” The presence of “inadequate” software training is subjective and is not defined or quantified within the standard. Machine learning models are trained by presenting the algorithm with sufficient data to generate patterns and predicted distributions for unseen data within an acceptable level of error. Engineers of highly automated agricultural machinery must select appropriate training data to represent their design specifications, or else the desired objective of the designer may not coincide with the actual learned pattern of the machine.
ISO 12100: Safety of Machinery–General Principles for Design–Risk Assessment and Risk Reduction
The ISO 12100 standard has been a widely applied standard for non-autonomous agricultural machinery used to improve the safety of a machine by presenting a recommended process for assessing risk. ISO 12100 considers the risk of a specific hazard to be a function of the severity of harm that can result from the hazard and the probability of occurrence of that harm (or situation that would lead to harm). The standard suggests using reliability, statistical data, and incident history to estimate the occurrence of a hazardous event. For HAAM, while the processes that are being automated are not novel, the introduction of sensors and artificial intelligence to substitute for human perception and logical reasoning is novel. Therefore, the use data, statistics, measures of incident frequency, severity, etc. would seem to be a flawed recommended approach, at least at this time, as such data do not yet exist. Though, it may be possible to examine data from non-autonomous machines to gain partial insights (Sandner, 2021).
Another key part of examining risk reduction in ISO 12100 is to examine and estimate exposure of a person to the hazard, including the need of a person to access the hazard zone to correct a malfunction or perform necessary repairs. As previously stated, ISO 18497 excludes farmyard or barn operations which are often locations for repairs to take place. Since the ISO 18497 explicitly references ISO 12100, this leads to a contradiction in scope, creating a gap in addressing the safety of humans during maintenance procedures in areas other than the farm field. An additional restriction that is created based on standard scope occurs within ISO 12100 which explicitly states that this standard, “does not deal with risk and/or damage to domestic animals, property or the environment.”
ISO 25119: 1-4 Tractors and Machinery for Agriculture and Forestry–Safety Related Parts of Control Systems
The ISO 25119 standard is comprised of a series of four parts that provide guidelines on: general principles for design and development (part one); the concept phase (part two); series development, hardware, and software (part three); and, production, operations, modification, and supporting processes (part four). A prerequisite for all parts of ISO 25119 is the completion of a, “suitable hazard identification and risk analysis (e.g., ISO 12100) for the entire machine.” The risk analysis technique specified in ISO 12100 follows the guidelines of a traditional Failure Modes and Effects Analysis (FMEA). The limits of ISO 12100 through its scope have already been noted. ISO 25119 introduces Hazard Analysis and Risk Assessment (HARA) as a related method of analyzing risk associated with faulted units. While FMEA has been a widely tested and validated risk assessment technique for a variety of automated systems, its efficacy when applied to highly automated agricultural machinery is limited because we currently lack suitable historical data on probability of failure occurrence or severity of incidents. Even if such data were available, it should be noted that quantifying probability and severity has always been at least somewhat subjective as and has been a weakness of FMEA as noted by Spreafico et al. (2017).
HARA in conjunction with FMEA serves to add a layer of redundancy to the risk assessment, performing a Hazard Analysis and Risk Assessment in accordance with ISO 25119-2 still requires the classification of avoidance of harm in relation to the ability of a human to control the situation. While HARA includes an evaluation of hazard “controllability,” FMEA asks the engineer to evaluate the “detectability” of a hazard. For HARA, levels and definitions controllability are shown in figure 2.
| C0 | C1 | C2 | C3 |
| Easily controllable | Simply controllable | Mostly controllable | None |
| The operator or bystander controls the situation, and harm is avoided. | More than 99% of people control the situation. In more than 99% of the occurrences, the situation does not result in harm. | More than 90% of people control the situation. In more than 90% of the occurrences, the situation does not result in harm. | The typical trained operator or bystander cannot generally avoid the harm. |
| Figure 2. Classification of avoidance of harm in accordance with ISO 25119-2. | |||
Classification of “avoidance of harm” in this manner would appear challenging in relation to highly automated systems because for highly automated agricultural machinery, where there is no direct operator, humans interacting with the machine have minimal knowledge of the status of the machine. To properly “control” the situation, a level of situational awareness is required to make an informed decision that will not result in harm. Additionally, there is a lack of data exploring both the training required to consider an operator “trained” to work with HAAM as well as exactly how humans and these automated machines will interact. Without a robust data set exploring the human/machine interaction during times of malfunctions or hazardous scenarios, there is no way to validate with any certainty what percentage of people would be able to control a given situation.
How a highly autonomous machine may react to a hazard is dictated both by the training the AI receives from the manufacturer, as well as the potential for a continuous process of learning it accumulates during normal operations. As such, the hazard reduction or avoidance maneuvers the machine may vary between manufacturers; leaving bystanders unsure of the subsequent actions the machine may take. This uncertainty removes the situational control from the human and places it with the machine. Using HARA as a tool to analyze machine-level malfunctions is further described in ISO 26262-3, however this standard was developed mostly for the automotive industry and is not entirely compatible within the scope of the agricultural industry.
Summary of Standards Analysis
The standards summarized in this project provide a reasonable baseline for analyzing risk associated with highly automated agricultural machinery. However, their efficacy at improving safety requires continued examination, refinement, and expansion as knowledge, experience, incident, or exposure data increases. It is possible that there exists room to build upon the existing standards to further improve their applicability to novel highly automated machines. The forthcoming revision of ISO 18497 is likely to advance the ability of engineers to assess risk much more effectively in the future. Some areas that may require development include the limitations in the scope of these standards leave critical areas (barns, farmyards, and public roadways) without explicit standard-based coverage. Additionally, much of the attention of recently developed standards have been on large-scale, self-propelled machinery such as tractors, combines, and harvesters. There exists the need for further conversation on safety surrounding the emerging designs and concepts such as smaller platform robots that might be used to feed livestock, maintain buildings, etc. as well of “swarm farming,” in which large numbers of smaller robots coordinate to perform a complex task as described by Anil et al. (2015). Finally, due to the recent release of a number of standards pertaining to HAAM, further research is needed to determine if industry leaders in automated agricultural machinery are using and applying the principles within these standards.
Survey Results
In February 2021, surveys were distributed to 706 individuals identified as potentially working with highly automated agricultural machinery. Of the distributed surveys, 173 were started, meaning that the Qualtrics link was clicked via a desktop computer or mobile device (smartphone, tablet, etc.). This represented an initial response (opening the survey) of 24.5%. The majority of initial responses followed the second and third mailings. Of those who initially clicked on the survey link, 128 completed the survey or 74% of those who initially began the survey. Of those 128, 56 respondents (44%) indicated that they have at some point designed or performed other forms of engineering work with highly automated agricultural machinery. Based upon the email requests from respondents to receive the “incentive,” the vast majority (over 95%) of respondents were from the industry sector, though to maintain anonymity, the specific work-related characteristics of respondents were not tracked. Further, a review of the full ASABE committee rosters reminds us that large-scale, multi-national manufacturers tend to be the most well-represented groups on various committees including those responsible for developing and promulgating standards, and this is also reflected in the responses—but again, IRB stipulations did not include identifying specific company characteristics to protect the privacy of respondents. Respondents were asked to specify the specific design standards they use in their work. Eight standards were presented as choices in the survey, with the additional option of writing in a non-listed standard. These were:
- ISO 12100
- ISO 25119-1
- ISO 25119-2
- ISO 25119-3
- ISO 25119-4
- ISO 18497
- ANSI/ASAE 318.18
- ANSI/ASAE 354.7
Figure 3 shows which of the standards are being used. Respondents were able to select multiple standards. The total number of responses selected for this question was 153 or an average of 2.7 standards per respondent. Note that half of respondents do use ISO 12100 which relies on FMEA (a method where additional future data is required for this method to be used effectively). Also note that the ISO 12100 standard’s scope excludes many of the exposures of concern noted in previous sections. ASABE/ASAE 318.18 is also widely used but does not explicitly cover issues of highly automated machines. Similarly, ISO/FDIS 18497, while a highly informative standard was used by almost one in three respondents, but in its current form only covers field operations, and it excludes highway and farmyard exposures.
 |
Figure 3. Standards used while designing highly automated machinery. |
Table 3 explores the differences between risk assessment techniques used for low or non-automated agricultural machinery and those used for highly automated agricultural machinery. These responses would likely be correlated to the responses connected to standards because some of the standards cited recommend specific techniques be used. Six methods of analyzing risk were presented to respondents: Failure Mode and Effects Analysis (FMEA), Fault Tree Analysis (FTA), Monte Carlo Analysis, Delphi Method, Informal Group Analysis (Brainstorming), and Risk Analysis Software.
In looking at the responses related to risk assessment technique, we noted that 12 respondents (21%) indicated they used a lower total number of risk analysis methods working with autonomous machinery as compared to the number of techniques they use with non-automated machinery. Overall, 39 individuals (70%) used the same number of techniques for both highly and non-automated machinery; though there was variation between which specific tools were used. For example, 33 individuals (59%) have used FMEA with non-automated machinery, however only 22 (39%) continued to use FMEA with highly automated agricultural machinery. Finally, among the 16 respondents who indicated that none of the formal risk assessment tools listed were used to examine highly automated machinery, nine of these respondents indicated they also used no formal tools listed with non-automated machinery. This means seven individuals used one of the listed risk assessment tools for non-autonomous machinery but did not use any specified technique with autonomous machinery.
| Table 3. Risk assessment techniques used in design for both low ornon-automated and highly automated agricultural machinery. | ||||
| Technique | N (low) | Frequency (%) (low) | N (high) | Frequency (%) (high) |
| Failure Modes and Effects Analysis (FMEA) | 33 | 59% | 22 | 39% |
| Fault Tree Analysis (FTA) | 12 | 21% | 11 | 20% |
| Monte Carlo Analysis | 9 | 16% | 8 | 14% |
| Delphi Method | 1 | 2% | 1 | 2% |
| Informal Group Brainstorming | 30 | 54% | 26 | 46% |
| Risk Analysis Software (various) | 6 | 11% | 5 | 9% |
| Other | 6 | 11% | 8 | 14% |
| None of the Above | 10 | 18% | 16 | 29% |
Techniques that were cited by individuals who selected “Other” while working with low or non-automated agricultural machinery included: Hazard Analysis and Risk Assessment (HARA), targeted product walk-arounds, functional safety assessments, and general risk assessment methods that are based on FMEA principles. Techniques written in by individuals who selected “Other” while working with highly automated agricultural machinery included: Hazard Analysis and Risk Assessment (HARA), targeted product walk-arounds, functional safety assessments, general risk assessment, formal discussion with a supervisor, legal reviews with an internal team, and industry seminars. Between levels of automation, more individuals referenced using HARA with highly automated machines as might be expected given that HARA is explained within the ISO 25119 standard series.
The final objective of the study was to examine the extent to which areas of risk are of interest and concern to survey respondents as they conduct risk analysis procedures or apply standards during their work designing highly automated agricultural machinery. We asked about specific exposures and resulting risk to humans operating machinery; bystanders; livestock; land/soil property; and civic infrastructures. These are shown in table 4.
| Table 4. Categories of exposure that respondents examine during a safety analysis of highly automated agricultural machinery. | ||
| Area of Consideration | N | Frequency (%) |
| Humans interacting directly with machines (e.g., operating, repairing, servicing) | 50 | 89.2 |
| Humans indirectly exposed to machine hazards (e.g., bystanders, motor vehicle operators) | 46 | 82.1 |
| Structures and built property (e.g., barns, storage units, buildings, homes, etc.) | 36 | 64.3 |
| Lost time/downtime that causes a production loss | 28 | 50.0 |
| Built civic infrastructure (e.g., roads, bridges, etc.) | 26 | 46.4 |
| Water (e.g., streams, rivers, ponds, lakes, groundwater) | 26 | 46.4 |
| Land/soil on the farm | 24 | 42.9 |
| Livestock | 24 | 42.9 |
| Non-agricultural wildlife | 17 | 30.4 |
| Land not owned by operator (adjoining or in close proximity) | 13 | 23.2 |
| Other | 3 | 5.4 |
In table 5, respondents were asked to describe how often they considered category of exposure as they conducted risk assessment with highly automated agricultural machinery. This question included a four-point Likert scale where a response of 4 was “always” and 1 was “never.”
| Table 5. Degree to which exposures to hazards are considered in risk assessment processes for highly automated machines. (4 – Always; 3 – Very Often; 2 – Sometimes; 1 – Rarely). | ||
| Exposures to Hazards | Number Citing “Always or Very Often” | Frequency Who Citing “Always or Very Often” (%) |
| Humans interacting directly with machines (e.g., operating, repairing, servicing) | 46 (of 50) | 92 |
| Humans indirectly exposed to machine hazards (e.g., bystanders, motor vehicle operators) | 44 (of 46) | 95.6 |
| Structures and built property (e.g., barns, storage units, buildings, homes, etc.) | 35 (of 36) | 64.3 |
| Lost time/downtime that causes a production loss | 26 (of 28) | 92.9 |
| Built civic infrastructure (e.g., roads, bridges, etc.) | 20 (of 26) | 76.9 |
| Water (e.g., streams, rivers, ponds, lakes, groundwater) | 18 (of 26) | 69.2 |
| Land/soil on the farm | 18 (of 24) | 75.0 |
| Livestock | 17 (of 24) | 70.8 |
| Non-agricultural wildlife | 7 (of 17) | 41.2 |
| Land not owned by operator (adjoining/close proximity) | 6 (of 13) | 46.2 |
Additionally, in the analysis of survey data, some demographic exploration was performed to examine at geographic location of respondents (their employer) as well as the age of the respondents. Of the individuals who worked on the design, application, and risk assessment of autonomous machinery, 87% were based in North America, 7% in Europe, and 2% each in Africa, Asia, and Australia. The number of respondents was small enough that we did not attempt to characterize variation in response differences by age (or geography) but note that most respondents were mid-career with 73% being were 41 years or older; 20% between 31 and 40 years old; and 7% were under the age of 31.
Discussion and Conclusion
This study has examined the availability, coverage, exclusions (through scope), and methods of risk assessment in existing standards that are used in the design of highly automated agricultural machines. This examination looked mainly at “field” based automated machines (as that is a primary focus of many standards and an area of high visibility), and we did not dive as deeply into smaller scale machines such as feeding equipment, barn cleaners, milking systems, or other automated, autonomous, or robotic machines used in buildings and on farmsteads. Note that new standards are being developed and are in various forms of draft stages. Most notably: ISO-CD 3991 Robotic Feed Systems – is currently listed as being “in development” at the time of this research. The standard ISO 20966 titled Automatic milking installations - Requirements and testing was published in its first edition in 2007 and contains substantive content on human safety. Both standards do reference ISO 12100, which again, recommends FMEA as a method for risk assessment, and recall that ISO 12100 excludes risk to animals. The limitations of each of these issues has been described in this article.
Regarding standards, it should be expected with any new technology that is advancing rapidly that standards and techniques for risk assessment will need to also change rapidly as engineers and others gain experience and data is collected that can better inform the risk assessment process. It has been noted that commonly referred to methods (FMEA, HARA especially) involve the identification of hazards or potential ways in which a component or system “failure” might occur and estimates of the probability of occurrence (or encounters with the hazard) as well as the severity of outcomes if that failure or hazard encounter were to occur. At this stage, having a robust dataset upon which to base these analyses is not really possible given the fact that these designs are novel and we do not yet have the data to fully inform a risk assessment process.
Additionally, as has been noted, many of the standards examined in this study exclude key hazards or situations where harm could occur. And they exclude many of the situations or targets of hazards (soil, water, animals, structures, etc.). But the survey showed that engineers are keenly interested in risk assessment and safety that goes beyond the human operator, obstacles in the field, etc. based on the frequency with which they responded on considering different targets of hazards during their risk assessment considerations. This is commendable and suggests that those involved in design of these machines are being thoughtful in their consideration of risk. Human safety is certainly important given the high rate of farm and ranch workplace injuries and fatalities, but other forms of risk and targets of hazards could influence the adoption and ongoing reputation of new forms of agricultural machinery.
As an example, a failure of an autonomous machine that results in a serious highway collision or a spill/release of a chemical into a major water source could strongly influence public perception, insurers, regulators, and other stakeholders. It would seem logical and likely that engineers are filling in gaps where no standards exist with techniques respondents identified in the survey like brainstorming, product walk-arounds, functional safety assessments, and others. It is also possible that existing standards that have exclusions through their scopes are simply being “stretched” in order to cover issues of concern despite the limits imposed by scope. These techniques and creative workarounds have value. One common element of good risk assessment is to use the collective imagination and experiences of teams of engineers (as well as other stakeholders including end users, sales staff, legal experts, safety specialists, and others) to envision all possible “failure” modes or potential misuses as they consider design solutions to mitigate risk regardless of their perceived probability. It would also seem valuable that designers and others involved in the development process have some degree of working knowledge about production agriculture, machine operation, worker behaviors, etc.
From the survey, we learned that the standards that are available are being used to some extent by practicing engineers and design professionals. Half of the respondents are using ISO 12100 which applies to all machine forms (highly automated and not) and that relies on FMEA. Almost half of those who responded are also consulting ASAE 318.18 Safety for Agricultural Field Equipment, an older standard not necessarily written to cover the types of automation we currently see in the industry.
As a general observation that stems from conducting the standards analysis and developing the survey, the entire practice of identifying and using standards and applying risk assessment methods (formal and informal) is complex and at times confusing. Standards are constantly changing and at any given time might be in the stages of initial drafting, revision, or other forms of transition. Scopes between standards can result in unintended exclusion of situations or conditions as they may refer to each other yet not be fully applicable to the hazard or situation of interest. It appears that industry personnel have a desire to look holistically at all forms of risk including human plus harm to soil, water, structures, animals, and the significant risk of failure that results in downtime. Standards are a vital part of the solution, but perhaps not the only answer.
We also did not look at this as part of this study, but cyber risk is not an insignificant hazard. Work by Drewry et al. (2019) showed that among farmers using various forms of digital technology including early forms of automation saw security and privacy as major concerns. This would seem to be an area ripe for future research.
Because of these complexities and the fact that full-scale automation in production agriculture is in its relative “infancy,” it seems other tools are needed that integrate standards, specific techniques (like FMEA and HARA), and point to appropriate data that can be used to conduct a robust risk assessment. As part of the work by Sandner (2021), pilot software was developed that incorporated robust and thorough information using a national database of past incidents with non-autonomous equipment that might be used to provide probability and severity estimates until new data can be collected over the coming years.
Additionally, from this survey and experience by the authors in the classroom and in seeing that 29% of survey respondents are NOT using any sort of the risk assessment methods we examined, more can be done in terms of classroom education for new engineers in ABET-accredited engineering programs. And perhaps more can be provided for the industry to help practicing engineers and technologists to conduct formal and informal processes for risk assessment through professional development, continuing education, and other efforts to build awareness and encourage adoption of new methods, especially since this survey showed many practicing engineers working on these technologies are in the mid to older-age demographics. In the process of bringing practitioners together, the interaction and conversations that can occur around issues of risk assessment can help create awareness and the sharing of ideas might help, especially in light of the gaps created through the current state of standard development and promulgation.
Another recommendation is to develop software or other decision support tools that consider all of the critical risk assessment needs embedded within existing standards and specific best practices in risk assessment, risk control, etc. to assist engineers. Such software would ensure all applicable standards are being fully considered and that all forms of risk (and the targets that could be harmed by hazards) are fully considered in ways that improve safety for all (including the end user, the general public, and the companies designing and selling new technologies). Software could also incorporate as much data as can be gathered on incident epidemiology connected to adverse events (deaths, injuries, and other forms of loss).
Through standards and tremendous attention being paid by the industry to improve safety, we know that non-automated machines have become much safer in recent decades, and the same is possible with new forms of highly automated equipment if we pay close attention to the issue and work on the basis of sound data and understanding.
Supplemental Materials
The supplemental materials mentioned in this article are available for download from the ASABE Figshare repository at: https://doi.org/10.13031/22048208.
Acknowledgments
This work has been supported by the USDA National Institute of Food and Agriculture, Hatch Project 1022349.
References
Anil, H., Nikhil, K. S., Chaitra, V., & Sharan, B. S. (2015). Revolutionizing farming using swarm robotics. Proc. 2015 6th Int. Conf. on Intelligent Systems, Modelling and Simulation (pp. 141-147). IEEE. https://doi.org/10.1109/ISMS.2015.26
ANSI/ASAE. (2017). S318.18: Safety for agricultural field equipment. St. Joseph, MI: ASABE.
ANSI/ASAE (2018): S354.7: Safety for farmstead equipment. St. Joseph, MI: ASABE.
Bahr, N. J. (2015). System safety engineering and risk assessment: A practical approach. CRC press.
Barnett, R. L. (2020). On the safety hierarchy and hierarchy of controls. Am. J. Mech. Eng., 8(2), 61-68.
Case IH. (2018). Case IH defines categories of autonomy and announces pilot program. Retrieved from https://www.caseih.com/emea/en-za/News/Pages/2018-02-14-Case-IH-Defines-Categories-of-Autonomy-and-Announces-Pilot-Program.aspx
Castillo, M. (2020). US farm employers respond to labor market changes with higher wages, use of visa program, and more women workers. Amber Waves: Econ. Food Farming Nat. Resour. Rural Am., 2020(1490-2020-1708).
Dash, S., Sarkar, S., Tripathy, H. P., Pattanaik, P., & Patnaik, S. (2021). Robotics in weed management: A new paradigm in agriculture. Proc. 2021 Int. Conf. on Electronic Information Technology and Smart Agriculture (ICEITSA), (pp. 561-564). https://doi.org/10.1109/ICEITSA54226.2021.00111
Dillman, D. A. (2011). Mail and Internet surveys: The tailored design method—2007 update with new Internet, visual, and mixed-mode guide. John Wiley & Sons.
Drewry, J., Shutske, J., Trechter, D., Luck, B. & Pitman, L. (2019). Assessment of digital technology adoption and access barriers among crop, dairy and livestock producers in Wisconsin. Computers and Electronics in Agriculture 165 104960. https://doi.org/10.1016/j.compag.2019.104960. (104; 1.06)
Freyhof, M., Grispos, G., Pitla, S., & Stolle, C. (2022). Towards a cybersecurity testbed for agricultural vehicles and environments. arXiv preprint arXiv:2205.05866.
Gerberich, S. G., Gibson, R. W., French, L. R., Lee, T.-Y., Carr, W. P., Kochevar, L.,... Shutske, J. (1998). Machinery-related injuries: Regional Rural Injury Study—I (RRIS-I). Accident Analysis Prev., 30(6), 793-804. https://doi.org/10.1016/S0001-4575(98)00032-3
ISO. (2007). 20966: Automatic milking installations—Requirements and testing. Geneva, Switzerland: ISO.
ISO. (2008). 26322-1: Tractors for agriculture and forestry—Safety - Part 1: Standard tractors. Geneva, Switzerland: ISO.
ISO. (2009). 10975: Tractors and machinery for agriculture - Auto-guidance systems for operator-controlled tractors and self-propelled machines—Safety requirements. Geneva, Switzerland: ISO.
ISO. (2010). 12100: Safety of machinery—General principles for design—Risk assessment and risk reduction. Geneva, Switzerland: ISO.
ISO. (2011). 26262-3: Road vehicles—Functional safety - Part 3: concept phase. Geneva, Switzerland: ISO.
ISO. (2012). 13849-2: Safety of machinery—Safety-related parts of control systems - Part 2: Validation. Geneva, Switzerland: ISO.
ISO. (2015). 13849-1: Safety of machinery—Safety-related parts of control systems - Part 1: General principles for design. Geneva, Switzerland: ISO.
ISO. (2018). 25119-1: Tractors and machinery for agriculture and forestry—Safety related parts of control systems - Part 1: General principles for design and development. ISO.
ISO. (2018). 25119-2: Tractors and machinery for agriculture and forestry—Safety related parts of control systems - Part 2: Concept phase. Geneva, Switzerland: ISO.
ISO. (2018). 25119-3: Tractors and machinery for agriculture and forestry—Safety related parts of control systems - Part 3: Series development, hardware and software. Geneva, Switzerland: ISO.
ISO. (2018). 25119-4: Tractors and machinery for agriculture and forestry—Safety related parts of control systems - Part 4: Production, operation, modification and supporting processes. Geneva, Switzerland: ISO.
ISO. (2021). 4254-1: Agricultural machinery—Safety - Part 1: General requirements—Amendment 1. Geneva, Switzerland: ISO.
ISO/FDIS. (2018). 18497: Agricultural machinery and tractors— Safety of highly automated agricultural machines—Principles for design. Geneva, Switzerland: ISO.
ISO/TR. (2012). 14121-2: Safety aspects—Guidelines for their inclusion in standards. Geneva, Switzerland: ISO.
ISO/DIS. (2022). 3991: Agricultural machinery—Robotic feed systems—Safety (in development). Geneva, Switzerland: ISO.
Kootstra, G., Wang, X., Blok, P. M., Hemming, J., & van Henten, E. (2021). Selective Harvesting Robotics: Current research, trends, and future directions. Curr. Robot. Rep., 2(1), 95-104. 10.1007/s43154-020-00034-1
Li, S., Aby, G. R., Aherin, R., & Issa, S. F. (2022). Farm fatal injury trends in Illinois from 1999 to 2019. J. Agric. Saf. Health, 28(2), 99-108. https://doi.org/10.13031/jash.14898
Lowenberg-DeBoer, J., Behrendt, K., Ehlers, M.-H., Dillon, C., Gabriel, A., Huang, I. Y.,... Rose, D. (2022). Lessons to be learned in adoption of autonomous equipment for field crops. Appl. Econ. Perspect. Policy, 44(2), 848-864. https://doi.org/10.1002/aepp.13177
National Safety Council. (2022). Work overview - NSC injury facts. Retrieved from https://injuryfacts.nsc.org/work/work-overview/work-safety-introduction/
Rainbow, R. (2021). Code of Practice - Agricultural mobile field machinery with autonomous functions in Australia. Grain Producers Australia (GPA), Tractor and Machinery Association (TMA) and the Society of Precision Agriculture Australia (SPAA). Retrieved from https://www.grainproducers.com.au/codeofpractice
Rakhra, M., Deb, P., Dahiya, O., Chandel, S. S., Bhutta, B., Badotra, S.,... Singh, D. (2022). An analytical study of the types of implements used by farmers in mechanized agriculture. Proc. 2022 Int. Mobile and Embedded Technology Conf. (MECON) (pp. 683-687). IEEE. 10.1109/MECON53876.2022.9751983
SAE. (2021). J3016_202104: Taxonomy and definitions for terms related to driving automation systems for on-road motor vehicles. Warrendale, PA: SAE.
Sandner, K. (2021). The efficacy of risk assessment tools and standards pertaining to highly automated agricultural machinery. University of Wisconsin-Madison. Biological Systems Engineering Department.
Spreafico, C., Russo, D., & Rizzi, C. (2017). A state-of-the-art review of FMEA/FMECA including patents. Comput. Sci. Rev., 25, 19-28. https://doi.org/10.1016/j.cosrev.2017.05.002
Wogalter, M. S. (2018). Hazard analysis and hazard-control hierarchy. In Forensic human factors and ergonomics (pp. 17-32). CRC Press.
Xu, W. (2020). From automation to autonomy and autonomous vehicles: Challenges and opportunities for human-computer interaction. Interactions, 28(1), 48-53. https://doi.org/10.1145/3434580
Zhang, Q., & Pierce, F. J. (Eds.). (2013). Agricultural automation: Fundamentals and practices. CRC Press.