Deep Neural Networks for Unsupervised Robotics in Building Constructions: A Priority Area of the Fourth Industrial Revolution (4IR)

Abstract

Many effective quality systems to maintain the robots’ autonomous task expansion process in construction industries for various applications over the years have yet to be well established. This study, therefore, presents a simple deep/neural network algorithm to diverse robotics tasks on building construction—bricklaying, grasping, cutting materials, and aerial robot obstacle avoidance and highlight the strengths of these algorithms in real-world robotics applications in building sites. Our findings revealed that the amount of tasks robots encountered in real-world environments is extremely challenging for existing robotic control algorithms to handle. Also, our algorithm when evaluated against other conventional learning algorithms can be a more powerful tool with the capacity to learn features directly from data, making it an excellent choice for such robotics applications in building construction. In other words, our algorithm can teach robots the ability to “work,” “think,” “know,” and “understand” their surroundings. It can also improve customer satisfaction, speed up the building process, and improve the productivity of building development teams. This chapter, however, contributes to classifications of autonomous robotics application development in construction literature. Although the problem addressed in this chapter is based on building construction, the algorithms presented are designed to be generalizable to related tasks.

1. Introduction

Robotics has influenced nearly every modern construction industry by improving efficiency, safety, and cost [1]. Robotics offers process automation and reliability thanks to sensor technologies [2]. However, while most industries have embraced this new technology from the moment it is released, many construction industries have historically been slow to onboard automated solutions [3]. The reason could be linked to product features and complexity (project size, lifetime and uniqueness, versatile construction environments). According to Ref. [4], several other phenomena could add to the characteristics of these construction industries that often contribute to the complexity of projects. This phenomenon includes client needs that are sometimes imprecise and changing, causing significant change in costs; little overall learning because of few repetitions; high risks due to novelty; technical, climatic, and even societal uncertainties; coordination and complex decision-making processes between the teams involved; and changing conditions of realization. The weak capital budgets in R&D and the reluctance of strategies related to construction automation are other important factors [5].

These phenomena have remained unresolved for a variety of reasons. First, building nowadays are complex entities, and construction entails many different trades coming together to work in perfect sync with each other [6, 7, 8, 9]. A replica of human-like dexterity, intelligence, and situational awareness, developed over hundreds of millions of years, is needed to break even [10]. Secondly, building construction project sites is often chaotic, disorganized spaces with materials, tools, debris, and wires spread about. Many areas of the sites have unpaved soft soil, into which a builder may sink if he steps off the beaten path. There are some environmental factors such as dust, rain, ice, and storms. There are humans walking around, etc. All these have not improved for so long because construction industries are the least digitized sector [3, 5].

Nevertheless, several ongoing initiatives suggest a gradual change in practices in this industry for rapid industrialization, which is enabled thanks to robotics and automation. According to Ref. [11], rapid industrialization is always based on quantity and quality that includes prefabrication, mechanization, automation, robotics, and finally reproduction. This gradual change toward rapid industrialization is driven by a concern to change the narrative and to be in line with the innovations observed at the international level that will respond to important building construction challenges: workforce, competitiveness, sustainable development, etc. In view of this, digital transformation with robotics technologies is one of the preferred avenues to improve the sector’s overall performance over the long term [12]. This digital transformation is expressed in different ways. One of the strong currents is based on the so-called concept of the Fourth Industrial Revolution (4IR), based on unsupervised robotics utilization in building constructions.

1.1 Rationale and research gap

While the robotics use in building construction will only continue to grow from traditional design through final inspection and maintenance, the full benefits of construction robotics have yet to be realized. For example, as robotics begins to move from the lab to the real world, robots face many new challenges. A building construction assistant robot, for instance, must perform many complex tasks such as bricklaying, foaming, sorting, operating appliances, picking up, and cleaning materials in the site. It must also handle the huge variety of objects, materials, and the likes associated with these tasks such as picking up different objects, some of which it may never have seen before. For all of these problems, there exists only an abstract relationship between the robot’s visible inputs and the task at hand.

Traditionally, a roboticist, or team thereof, would hand-design these robots for each task they want the robots to perform. Even for tasks which human users can perform intuitively, such as bricklaying, grasping, cutting objects, detecting, and avoiding various forms of obstacles in building construction sites, and these robots can be very difficult to design because developers were not able to easily translate these abstract intuitions between the robot’s visible inputs and the task at hand into code. This makes it extremely challenging to scale these approaches up to the huge amount of works and obstacles that building construction robots must deal with in the real-world construction.

Secondly, most robotics that engaged in construction activities primarily rely on a system performance metric that is dependent on a metric connected to the given human-defined tasks, or task-dependent metric. This also implies a conventional paradigm of same manual development, where human designers were in charge of planning and coding for particular tasks-based activities that the robot would perform and how it would perform them [13, 14, 15]. This has some drastic limitations on the construction industry. For example, it is impossible for a bricklaying robot to perform delicate tasks such as installing electrical cables, or even to be able to detect and avoid various types of obstacles on the building construction sites. This limitation caused during design and coding thus prevented these robots from independently “thinking,” “knowing,” or “understanding” the multiple building operations thus generating questions about what developers had overlooked that made them to fail to get the anticipated outcome of operational autonomy.

Several researchers have criticized this conventional architecture for its lack of computational data [14, 15, 16, 17, 18, 19] though went on to propose several failed models, which would have addressed this research gap. For example, in their research, several researchers [14, 15, 16] provided the first implementations of this autonomous curiosity, but they were unable to integrate their concept within the issues with construction of robo-mason by demonstrating how the robot’s work patterns could emerge without the assistance of a human.

However, in order to establish this brain model that will give the robot significant cognitive developing abilities unique to humans, this study plans on modeling the brain at a level above the neural level, or what would normally be thought of as the unsupervised or unmanned level using neural deep learning algorithms that will allow the robots to learn independently from some training data. Our understanding of brain abstraction is sufficient to program a system that exhibits similar properties and connections to the human brain without having to model its detailed local wiring. Quite clearly, we will model this machine learning algorithmic concept based on neural networks and highly-parameterized models, which will use multiple layers of representation to transform data from a task-specific representation to an autonomous task. While our algorithm is not designed to solve central problems in artificial intelligence (AI), such as speech or object recognition, its development was motivated by the need to improve the performance of robots in the building construction industry. Traditional algorithms often struggle to generalize well on AI tasks specific to this domain, which prompted us to explore the potential of deep learning.

This chapter presents a distinct application of deep neural learning algorithms that can enable a building construction robotics to learn from some training data and to perform highly cognitive artificial intelligence operations. Here, rather than forcing the engineer to hand-code an entire end-to-end construction robotic system, our machine learning algorithm will allow portions of the system to be learned from some training data. This approach will allow us to model concepts, which might be difficult or impossible to properly hand-model. It will also allow for adaptable models—as long as the form of the model is general, meaning it can be adapted to more or different cases simply by providing training data for these new cases.

However, by using unsupervised feature learning algorithms, deep learning approaches are able to pre-initialize these networks with useful building construction features, thus avoiding the overfitting problems commonly seen when neural networks are trained without this initialization. This machine learning algorithm can therefore work very well on a wide variety of construction projects.

Nevertheless, we begin with a general description of deep learning algorithms for unsupervised feature learning as well as their strengths and particular advantages as learning algorithms for robotics applications on building constructions. Finally, we present a simple deep/neural network algorithm to diverse robotics tasks on building construction—bricklaying, grasping, cutting materials, and aerial robot obstacle avoidance—highlighting the strengths of these algorithms in real-world robotics applications in building sites. It is our hope that these algorithms will demonstrate a more appropriate computation model where, for instance, robots’ artificial intelligence and ability to detect obstacles and carry out multiple construction tasks unsupervised are no longer isolated from the subjective experience of the body.

2. Literature review

2.1 Deep learning/neural network for unsupervised robotics technologies

Deep learning is a machine learning approach based on modeling adaptation of biological neural systems. It can also be defined as a computing system made up of a number of simple, highly interconnected processing elements, which process information by their dynamic state response to external inputs [20]. Information processing is carried out through connectionist approach to computation and this amount of information needs a complex abstraction as data representations through a hierarchical learning process [21]. The term hierarchical learning here is referred to neural networks [22]. An example of such a neural network is shown in Figure 1.

Figure 1.

Multi-layer artificial neural network.

Here, our algorithm could be implemented to train the network in an unsupervised manner. This is because the multi-layer (with many hidden layers) neural network is being used as shown in Figure 2 in which each layer takes input from the previous layer, processes it, and outputs it to the next layer, in a daisy-chain fashion.

Figure 2.

Deep neural network for deep learning.

With this, our deep learning can be used to generate high level of abstraction for the building construction robots. So for complex abstractions of data representations through a hierarchical learning process, our deep learning model can produce results faster than standard machine learning [23]. And also our proposed deep learning model will be integrated with building construction features that are important by itself to be learned, instead of requiring to be manually selecting the pertinent features. However, this unsupervised learning model does not require the presence of a teacher. The desired output is not presented to the network. The system learns on its own by adapting the structural features in the input patterns. The general role of our unsupervised learning model is shown in Figure 3.

Figure 3.

Classification and clustering.

2.2 Robotics application and its capacity on building construction

Robotics applications in construction are a large field of study due to the multidisciplinary trades that constitute the act of constructing. For example, bridges are a type of construction subjected to various robotic automations. Oh et al. [9] developed a robot for bridge inspection and Lorenc et al. [24] for maintenance. Bridges can be difficult to access, and there is a high demand for robots to perform diagnostics and repairs. However, we have gathered information from various resources especially from relevant literatures during a span of 10 years, i.e., 2013–2023 pertaining to more forms of robotics’ applications and its capacity in building construction.

2.2.1 Drones/UAV technologies application in building construction

Few numbers of references are reported in the literature regarding use of drones/UAVs technologies in construction by many authors [3, 25, 26, 27]. The reports show that the use of drones highlights the essential role of humans in robotics, as drones are being utilized for a range of tasks from painting to identifying safety concerns on work sites. See Figure 4. These robotic manipulators are appropriate for the different spraying activities involved in construction work. For example, high-rise building coating, fireproofing application, and shot blasting can all be done without the need for a human operator to be physically present. With the aid of drone robots, a variety of applications, including cleaning and sandblasting, may be completed effectively. Drones that paint walls in tall buildings are among the robotic technologies. It comes equipped with automated spraying equipment and shot blasting capabilities for surface preparation.

Figure 4.

Drone in construction work site.

Also, drone data has been integrated with data captured by ground-based robots, known as Autonomous Mobile Robots (AMRs), to provide a real-time view of a work site [28]. This data can then be integrated into virtual reality, allowing project managers to see the construction site without having to leave the office.

Drone data becomes necessary because traditional inspection is labor-intensive. Due to the harsh workplace and lack of trained personnel, site inspection can be completely automated using feedback from auto survey tools. These tools have a video and electronic display that transmits data to the microcomputer using fiber optic cable and employs a laser to configure the shield machine. This drone technology has been briefly applied in Architecture, Engineering, and Construction (AEC) domain and provided guidance for UAV operation and implementation in the construction industry [29]. Scaled Robotics are forms of drone which are developed as Autonomous Mobile Robots (AMRs) that are controlled via mobile devices where information is collected by robots using laser scanners, focusing on identifying obstacles, mistakes, or errors in an effort to reduce rework [30]. This is done with the aid of some digital photogrammetric systems that are combined with robots for demolition and site cleanup to manage large-scale earthwork projects. They are also remote-controlled vehicles featuring microwave communications, radar beacons for position, and radar sensors for avoiding obstructions and identify errors. They are, therefore, capable of controlling up to 100 units simultaneously and can identify impediments at a distance of up to 2 kilometers.

Again [31, 32], extensively conducted a systematic literature review on the current topic pertaining to implementation of Unmanned Aerial System (UAS) in the construction industry covering the most relevant job, cases, and areas of application. Their report revealed that the recent developments in UAS regulations have played a significant role in their popularity and wide deployment in various stages of the construction lifecycle. A comprehensive study conducted in the United States to identify the practical construction UAS application areas, their adopted technologies, as well as the benefits and barriers encountered during their implementation [33], and further revealed that drone robotics technologies today help many aspects of the building construction industries, including knowledge-based design and control [34].

However, a critical literature review has also been carried out on the relevant existing studies toward this immersive and digital technology [6]. The authors analyzed the literature using meta-synthesis technique to evaluate and integrate the findings in a single context. Their review shows that this class of construction drone robots for various purposes focuses mostly on tasks in hazardous areas where people cannot perform them. These domains include deep sea oil prospecting, nuclear power plants, and decommissioning issues. However, counter reports have revealed that these immersive and digital technologies have not been significantly utilized in the construction sector and by extension of other sectors of the economies in developing countries due to the high level of investment required [35, 36, 37, 38]. It therefore makes sense to utilize these technologies that can function without administrative or motivational problems in hazardous conditions, bad weather, and at night to reduce the growing number of accidents in hazardous building tasks [39].

2.2.2 Bricklaying robots

Bricklaying work is one of the most arduous jobs in construction [40, 41] since it includes a mason standing, kneeling, and lifting. In addition, the mason works almost exclusively outside and undergoes the weather conditions (rain, wind, heat, humidity). The mason sometimes works in height scaffolding or in trenched soils, which may put his life in jeopardy. However, in the last two decades, some research projects focused on the development of a bricklaying robot [42]. Bricklaying work follows predefined steps and thus is favorable for automation. However, the process cannot be fully automated and requires the supervision of a worker nearby to adjust/control the robot. Tan et al. [43] stressed the importance of the environment when designing a robot. They support the idea that robot level of autonomy should be in line with the environment (actively/passively/not assisted environment). For that, the authors proposed a framework to help categorize the robot/environment interaction.

Recent advancements in masonry work automation technology include Australian Hadrian X and Construction Robotics’ SAM100 also known as semi-autonomous mason are robotic bricklaying machines [44, 45]. Hadrian X uses an intelligent control system alongside Computer Aided Drawing (CAD) to function and is capable of building a standardized home every two days on average. The robot is capable of laying the bricks with a high accuracy thanks to a laser guidance system. It is also able to work on almost any block size. The advantage of such a design is the flexibility in mobility: The robot can work under difficult circumstances linked to the environment. That is to say, by deploying Hadrian on a construction project, one can benefit from faster masonry work, least material wastage, and overall cost-efficacy. Hadrian X closely resembles a truck crane (Figure 5).

Figure 5.

Bricklaying robot “HADRIAN X.” source: Pivac and Pivac [44].

SAM100 (Semi-Automated Mason) on the other hand is designed to work collaboratively with masons to increase productivity by 3 to 5 times while reducing lifting by 80% and so on [45]. The robot has successfully passed the prototyping phase and is now commercially available. Figure 6 shows the utilization of SAM100 onsite. This robot is by far the most complete masonry robot realized until now. It can lay bricks with precision and includes the binder in the process of laying as well. SAM100 is capable of laying 800–1.200 bricks a day. The robot performs in a straight line with a limited height capacity. SAM100 costs around 500.000$ (442.030 €).

Figure 6.

SAM100 robot onsite screenshot. Source: Podkaminer and Peters [45].

2.2.3 Specific design building construction robots

Several other research projects have focused on a specific design problematic of the automation process. For example, SAM100 design is based on an articulated arm as found in previous research projects [12, 46]. The “HADRIAN X” is based on a variant of the articulated arm supported by a truck-crane robot. FUNAC robots and COGIRO robots with long arms were also not left out in the discussions. COGIRO robot is used as a precise tele-operated crane to position prefabricated roof elements, while the FUNAC robots does monotonous, risky, repetitive construction works, handling payloads up to 2300 KG due to the strength of its axis, see Figure 7.

Figure 7.

FUNAC robots with long arms placing bricks and other materials.

These robots promise to reduce operating costs and waste, as well as provide safer work environments and improve productivity. However, while deploying robots like the FUNAC in a building construction project, the number of axes it has must be a crucial factor to be considered [39]. Many people might not be aware of the significance of the robots’ axes and how they regulate a robot’s range of motion and strength of work. Each axis of a robot stands for a degree of freedom, or to put it another way, an independent motion that enables a construction robot to become more functional. In other words, the more degrees of freedom and higher usefulness a robot has, the more axes it has. For instance, six-axis FUNAC robots are perfect for repetitive, tiresome, and even dangerous construction tasks that were previously solely performed by people because designers mirror human arm movements and offer the same degrees of freedom as human arms [47].

FUNAC robots can therefore do whatever their human counterparts can do in building construction, without becoming fatigued or risking their safety. They have better access to even challenging vocations due to their wider range of motion and numerous degrees of freedom, thus they carry out several repetitive operations. It is perfect for the majority of applications found on construction sites, grasping, and handling payloads up to 2300 KG and reaching beyond 3.5 meters. In Figure 8 below, FUNAC robot models are displayed.

Figure 8.

The FANUC M-10ia and the FANUC R-2000ib six-axis robots.

These six-axis robots offer more advantages than other types since they have a wider range of motion and can move in more directions than just the x, y, and z planes owing to their various degrees of freedom. For example, a three-axis robot can only move in three planes (x, y, and z) because it lacks the other three axes. Robots with four or five axes may move in all three planes and can even perform extra actions like rotating or elevating the mixers.

However, due to the minimal amount of movement required to remove something from a conveyor and place it on a pallet, four-axis robots are frequently utilized in palletizing applications. Although practically any palletizing operation may be completed using the four-axis FANUC M-410ib/160 without the use of the 2 extra axes, it has been previously shown [39] that a robot’s range of motion increases with each axis it possesses.

In sum, we have conducted this extensive literature work to raise the consciousness of our readers on what the literature said regarding the applications and strengths of robotics in building construction. Our conclusion is that when applying robotics in building construction, engineers have a lesser work envelope because these machines are more versatile and have a wider range of work alternatives.

2.3 Unsupervised robots in building construction: Requirements and important qualities

2.3.1 Robot requirements for building construction tasks

• It should be effective in preventing human operators from dying in dangerous circumstances.

• It ought to function in dangerous conditions, in the dark, and without issues with administration or motivation that would be profitable [48].

• The benefits should be maximized across a variety of application areas.

• It must be mobile, autonomous, and cognitive.

2.3.2 Important qualities needed in construction robots

Sensing and control, mobility and manipulation, human aspects and task factors, expert systems, and task flexibility are some of the qualities. The following sections describe these qualities in more detail.

2.3.2.1 Sensing and control

The largest challenge in developing construction-related robots is sensing and control, particularly in terms of navigation and position. The mobile autonomous robot requires location and heading data constantly for control. With the aid of video and image recognition systems, obstacles can be avoided and objects can be located. On numerous prototypes, obstacle avoidance is accomplished using touch sensors and ultrasonic technology [49].

2.3.2.2 Mobility and manipulation

The ability of equipment to move about construction sites is influenced by a number of variables, including the types of working environments and surface materials that must be traversed. For example, robots installed on rails have enough mobility to perform a variety of finishing activities and wall inspection jobs [50]. Different robots will do manipulative jobs according to their load bearing capabilities, arm length, and grip style.

2.3.2.3 Human aspects and task factors

The main motivation for the development of construction robots for use in severe settings, high and deep locations, boiling seas, and radiation zones is safety [51]. In these applications, the robots could operate alone or in tandem with a human operator working remotely from a secure location. Man or machine must have overall control for safety reasons. However, it is important to consider the current state of telecontrol to prevent delay issues when dealing with challenging tasks and to provide accurate manipulation feedback to the operator. During teleoperation, human factors are very crucial, even though automation reduces manual labor, it often requires more mental and cognitive effort. At this stage, the human-machine interface is particularly important for machine control and display.

2.3.3 Expert system and task flexibility

The construction industry will not benefit much from robotics on its own. Real progress can only be made when the construction process is completely organized. Expert systems, CAD/CAM, and database technologies are crucial in robots for task flexibility in civil engineering applications [52]. This field is now conducting in-depth research on unique programming environments, engineering graphics, logic, computation, and control requirements. The loadings, material properties, components, connectors, assembly, and geometric reasoning system all play a role in how construction components are represented in three dimensions by an expert system. A computerized work-control system will be necessary for robotic work to be employed efficiently. This system must include the construction site organization and sophisticated processing of commodities from the site to the robot.

2.4 Theoretical framework and hypotheses formulation

The Cognitive Behavioral Therapy (CBT), which was developed by psychologist Albert Ellis in the 1960s, and the theory of neural network, which was developed by Levin and Narendra in 1993, have been employed to guide this study. The principles of cognitive behavioral psychotherapies have described how a robo-mason can recognize, interact, think, and give meaning to situations, detect obstacles as well as form beliefs about themselves, their environments, and the world. This theory serves as a mechanism for robots to interact with its environments in an appropriate manner based on its cognitive behavior [53]. This theory better explains our study because the conventional robot design philosophy has given considerably more consideration to the embedded structure than to cognitive autonomous behavioral concerns, despite the fact that both play a substantial impact in the behaviors that follows. Autonomous cognitive design considerations of construction robo-mason are now essential in this study for a robot with embedded structure to learn and develop in order to eventually adapt to increasingly complicated building construction environments [54]. The major advantage is for the robot to change from dependent, excessive, and unhelpful behavior patterns to autonomous, advantageous, and balanced solutions.

The second theory is the theory of nonlinear dynamic system using neural network developed by Levin and Narendra [55]. The theory explains the use of neural networks to improve the stability and controllability of robotic systems. Though their study is restricted to nonlinear systems with complete state information access and feedforward MLNs with dynamic BP, their method takes into account a discrete-time system at index k, as indicated in Eq. (1). The proposed design of the neural networks is shown in Figure 9.

Figure 9.

Proposed architecture of the neural networks in the work of Levin and Narendra [55].

$$\mathrm{x}\left(\mathrm{k}+1\right)=\mathrm{f}\left(\mathrm{x}\left(\mathrm{k}\right),\mathrm{u}\left(\mathrm{k}\right)\right)$$E1

where x(k) ∈ χ ⊂ Rⁿ, u(k) ∈ U ⊂ R^r and f(0,0) = 0 so that x = 0 is an equilibrium. Eq. (2) shows the conditions required for the neural networks to achieve feedback linearization and stabilization of the system.

$$\left\{\begin{array}{c}u={\mathit{NN}}_{\mathrm{\psi }}\left(v,z\right)\\ z={\mathit{NN}}_{\phi }\left(x\right)\end{array}\left(z,v\right)=\epsilon R^{n}x{R}^r\right.$$E2

However, Sontag tested and used this model to examine the potential and ultimate constraints of alternative neural networks designs [56]. He asserts that nonlinear systems like robots in general may be stabilized using neural networks with two hidden layers. Their conclusion seems to go against neural networks approximation theories, which contend that neural networks with a single hidden layer are the best approximators. In contrast to approximation problems, Sontag’s solutions are based on the representation of the control problem as an inverse kinematics problem.

3. Method

3.1 Building deep learning algorithm for unsupervised robotics for construction industries

In building deep learning algorithm for unsupervised robotics for construction industries, we described as particular instances of a fairly simple recipe: combine a specification of a dataset, a cost function, an optimization procedure, and a model. For example, the linear regression algorithm combines a dataset consisting of x and y, the cost function

$$\mathrm{J}\left(\mathrm{w},\mathrm{b}\right)=-\mathrm{E},\mathrm{p}log\mathrm{p}\left(\mathrm{y}|\mathrm{x}\right)$$E3

However, the model specification p (y|x) = N (y; x w + b, 1), and, in most cases, the optimization algorithm defined by solving Eq. (3) for where the gradient of the cost is zero. By recognizing that these components can be replaced independently, a broad range of algorithms can be obtained. The cost function typically includes at least one term that causes the learning process to perform statistical estimation. Thus, the most common cost function is the negative log-likelihood, so that minimizing the cost function causes maximum likelihood estimation. The cost function may also include additional terms, such as regularization terms. For example, we can add weight decay to the linear regression cost function to obtain

$$\mathrm{J}\left(\mathrm{w},\mathrm{b}\right)=\mathrm{\lambda }\Vert \mathrm{w}\Vert -\mathrm{E},{}_{\mathrm{p}}log\mathrm{p}\left(\mathrm{y}|\mathrm{x}\right).$$E4

This still allows closed-form optimization. For example, if we change the model to be nonlinear, then most cost functions can no longer be optimized in closed form. This requires us to choose an iterative numerical optimization procedure, such as gradient descent. The recipe for constructing a learning algorithm by combining models, costs, and optimization algorithms supports both supervised and unsupervised learning. The linear regression example shows how to support supervised learning. Unsupervised learning can be supported by defining a dataset that contains only x and providing an appropriate unsupervised cost and model. However, we can obtain the first PCA vector by specifying that our loss function is

$$\mathrm{J}\left(\mathrm{w}\right)=\mathrm{E},{}_{\mathrm{p}}\Vert \mathrm{x}-\mathrm{r}\left(\mathrm{x};\mathrm{w}\right)\Vert$$E5

while our model is defined to have w with norm one and reconstruction function r(x) = w xw. In some cases, the cost function may be a function that we cannot actually evaluate, for computational reasons. In these cases, we can still approximately minimize it using iterative numerical optimization so long as we have some way of approximating its gradients. Most machine learning algorithms make use of this recipe, though it may not immediately be obvious. If a machine learning algorithm seems especially unique or hand-designed, it can usually be understood as using a special-case optimizer. Some models such as decision trees or k-means require special-case optimizers because their cost functions have flat regions that make them inappropriate for minimization by gradient-based optimizers. Recognizing that most machine learning algorithms can be described using this recipe helps to see the different algorithms as part of a taxonomy of methods for doing related tasks that work for similar reasons, rather than as a long list of algorithms that have separate justifications.

4. Discussion

These feature learning approaches are one of the major strengths of modern deep learning methods. Since these algorithms are able to learn good features from data, they are much less sensitive to input representations than other conventional learning algorithms such as support vector machines, Gaussian processes, and others. Deep learning algorithms are able to learn good representations and solve problems even from basic representations such as raw pixels, avoiding the need to hand-design features as with other learning algorithms, saving significant engineering effort for many of the complex problems encountered in robotics, where features can be unintuitive and hard to design.

5. Conclusion

The branch of computer science that is now permeating the construction sector is robotics. It is crucial to automate construction sites so that robots can perform risky tasks for workers in dangerous environments like high altitudes, deep water, high radiation zones, inclement weather, and deep oceans. It is also beneficial in terms of avoiding the disruptive effects of strikes, issues with administration and motivation, safety and health regulations, a lack of skilled labor, and the need to perform repetitive, dirty, and dangerous work as well as the completion of projects or tasks with quality control, on schedule, and economically. Despite the fact that this technology is helping the construction industry, much research is still needed in the areas of sensing and control, human factors, task flexibility, and the software support to integrate robots into a larger construction-based management. Our feature learning approach is one of the major strengths to achieve this. Since this algorithm is able to learn good features from data, they are much less sensitive to input representations than other conventional learning algorithms such as support vector machines, Gaussian processes, and others. Our deep learning algorithm is able to learn good representations and solve problems even from basic representations such as raw pixels, avoiding the need to hand-design features as with other learning algorithms, saving significant engineering effort for many of the complex problems encountered in robotics, where features can be unintuitive and hard to design.

Acknowledgments

The authors express their gratitude to all that helped in the actualization of this research.

Conflict of interest

No potential conflict of interest was reported by the author(s).