The agri-robotics community’s recent focus has been on identifying applications where repetitive task automation is more efficient or effective than a traditional human or large-scale machine approach. Research is needed on robotic platforms that can operate near the crop (either on the ground or at elevation) and advanced manipulation, particularly with interactive or tactile properties such as soft fruit picking. The use of heterogeneous “multi-modal” platforms that combine ground-based and aerial vehicles provides the individual platforms with opportunities for targeted support and intelligence, plus the ability for observation and mission planning for human operators to have an “eye in the sky.” For large-scale arable and fruit crops, collaborative and cooperative behavior becomes advantageous as tasks can be performed in parallel, resulting in economies of scale. Land management is a particular issue of concern in the UK landscape, given the issues of soil fertilization, water management and carbon content, so it will become increasingly important to use advanced sensing and soil management using remote platforms including robotics. In addition, the use of robotics to manage livestock is a specific opportunity for the deployment of autonomous platforms, as has already begun in automated milking stations and with potential applications for animal husbandry in fields, barns, sheds and aquaculture or fish farms.
There are a multitude of related areas already using automation (such as large parts of UK industrial food production), and research is needed to investigate how this can be more closely integrated into the agricultural industry. The food chain is managed using complex food production and software systems that rely on accurate data on all aspects of agricultural food location, quality and quantity. In the processing side of the food industry, robotics and automation are already being used extensively, but this is not being leveraged to the same extent in the field on the production side. The use of large data sets in combination with remote sensing (as used in industrial production raw material tracking, e.g. with RFID tags) to optimize the quantity and quality of crops or livestock produced has the potential to revolutionize the UK agricultural sector. Technologies from related areas, including the Internet of Things, Big Data and Artificial Intelligence, can be used in conjunction with autonomous systems technologies to automatically fuse and interpret collected data, assess crop status, and automatically plan effective and timely interventions in response to sudden events and changes in crop conditions (e.g. weather, diseases, pests).
Our long-term technology vision includes a new generation of smart, flexible, robust, compliant, interconnected robotic systems that work seamlessly in farms and food factories alongside their human employees. In addition to and within existing Agri-Food systems, teams of multi-modal, interoperable robotic systems will organize themselves and coordinate their activities. Electric farm and factory robots with interchangeable tools, including low-tillage solutions, new soft robotic grasping technologies and sensors, will support sustainable agricultural intensification and drive productivity across the food chain. In order to increase their own productivity, future agri-robotic systems will deploy artificial intelligence and machine learning techniques. Meanwhile, research into alternative food production systems, including innovations from areas such as vertical agriculture, will also help address sustainable agriculture intensification while protecting the environment, food quality and health. The clear demonstration of economic benefits, which has always been the primary driver of change for the agricultural community, is a vital aspect of making this transition effective. Facilitating the transition to automation While full automation is often hailed as the ultimate goal in technological development, and future farming systems may look very different from today’s ones, only very few large firms can afford full automation disruption. It will require a gradual transition from current farming practices to achieve this long-term vision, and most farmers will need technologies that can be introduced step by step alongside and within their existing systems. Moreover, while some emerging robotic technologies already achieve or approach the robustness and cost-effectiveness needed for real-world deployment, other technologies are not yet at this stage. Soft fruit picking, for example, still requires basic sensing, manipulation, and soft robotics research. Thus, human and robotic collaboration is fundamental to increasing productivity and food quality, at least in the short term. A number of relatively low-cost platforms are now available that are certified for use in conjunction with human workers. However, work is required to identify the nature of the necessary robot-human interactions and joint workflows. Thus, much of the research needed in the short- and medium-term should focus on facilitating the “transition to automation”, with mixed systems likely to dominate in the coming years, benefitting from collaboration between humans and robots, with combinations of electric, diesel and hybrid powered vehicles, as the required technologies for electrification mature and become ready for market. In the short-term, progress may depend on retrofitting of existing farm vehicles. For example, a human-driven tractor could tow a variety of robotic implements for different field operations such as selective harvesting or weeding. In the longer term, autonomous robotic vehicles will start to replace the legacy vehicles. This trajectory will also enable the UK vehicle and implement manufacturers to develop new products that span the transition from the current diesel-powered farm vehicles to the robotic farming systems of the future. The UK is well placed to implement these changes due to its strong automotive sector in industrial and agricultural vehicles, with extensive infrastructure already in place.
Small, smart, interconnected, light machines
One advantage of modern robotics is its ability to build on low-cost, lightweight and intelligent components. Because of their prevalence in consumer electronics such as mobile phones, gaming consoles and mobile computing (laptops, tablets, etc.), high-quality cameras and embedded processors can be installed at very low cost on many platforms. New materials and manufacturing techniques such as additive manufacturing and advanced composites also make robotic platform manufacturing and deployment much cheaper and decoupled from a mainstream manufacturing or supply chain. A 3D printer located on a farm, for example, could be used to produce spare parts on demand, at very low cost, or even to improve the platform by adapting to local conditions. Furthermore, the use of collaborative and cooperative behavior in a robotic fleet offers the opportunity to spread tasks across multiple platforms and thereby reduce the damage caused on the soil or existing crops by heavy conventional agricultural platforms. Multiple data sources can also be used by the robotic fleet to calibrate the task, reduce waste and focus on areas of greatest need, potentially reducing fertilizer costs and impact on the environment.
ENABLING TECHNOLOGIES FOR FUTURE ROBOTIC AGRICULTURE SYSTEMS
A wide range of technologies will allow agricultural robotics to be transformed into the field. Some technologies will have to be developed specifically for agriculture, while other technologies already developed for other areas, such as autonomous vehicles, artificial intelligence and machine vision, could be adapted to the agricultural domain. Here we briefly review the current status, opportunities and benefits of various hardware-to-software enabling technologies, multi-robot systems, and human-robot systems.
Agricultural platforms can be divided into domain-and task-specific robots designed to perform a specific task on a given crop in a predefined domain and generic platforms designed to perform multiple tasks in different domains. It is likely that both will play important roles. Since farms have very different infrastructure in general, early robots can only operate on a given farm and only to a limited extent on different farms. Therefore, similar to current farm vehicles, we can see a combination of robots adapted to a specific task and the emergence of multi-purpose robots capable of performing a multitude of different tasks, analogous to the myriad of modern tractor usage cases. A common challenge is that for real-world conditions such as mud, rain, fog, low and high temperatures, most current robotic platforms are not robust. For example, most current manipulators in glasshouses are not equipped to handle moisture. Mechatronics and electronics Rapid prototyping techniques and low-cost processors have resulted in an explosion in the use of 3D printing and “maker” technology, increasing the potential of low-cost robotic platforms for a variety of applications. Using embedded software enables highly configurable and application-specific platforms that can be adapted to a variety of roles using common hardware modules. While such approaches have been widely used in UAVs and smaller-scale robots, there is considerable scope for robotics expansion in Agri-Food on a much wider scale. Issues that need to be addressed in order to migrate from prototypes to robust commercial platforms include robustness and reliability, power management (platforms need to be able to operate all day long, in some cases 24/7), usability (non-specialists must be able to use platforms effectively), maintenance (e.g. self-diagnosis) and integration with mobile communications. Other challenges include better characterization of the soil’s mechanical properties relevant to these robots, rugged platforms capable of operating in inclement weather, real-time sensing and control algorithms to adapt locomotion strategies to an ever-changing environment, and locomotion co-design with other capabilities. For instance, how does the collection of crop / fruit affect locomotion? What locomotive capacity do we need to enable effective crop detection?
Agricultural robots need to move in challenging dynamic and semi-structured environments. Ground robots needs to travel on uneven, inhomogeneous, muddy soil, while aerial vehicles need to operate for long periods of time, in different weather conditions. Current agri-robots are mainly designed by borrowing technology from other sectors (e.g. drones) or as an add-on to existing platforms (e.g. autonomous tractors). As such, they may be not fully optimised for their tasks, or may retain some of the limitations of existing platforms.
UAVs can fly using multiple rotors or a fixed wing platform (with precision of location in the former and extended flight time in the latter), whereas ground platforms need to be able to locomote on rails and concrete floor in greenhouses, on gravel or grass in polytunnels, and in extremely muddy and difficult terrain in open fields . We will therefore see a wide variety of robots being developed with different means to locomote. Compared to tractors, these robots are extremely lightweight, but as robots (or autonomous tractors) are to perform more energy-demanding task, the robots will also increase in size and weight. Most agricultural robots today run on batteries and electrical motors. Future developments will depend on how the battery technology evolves, but we will probably see both electric and combustion engines in the field for the foreseeable future.
A key aspect of any robotic platform is the impact of the weight and locomotion system on the ground and crops, and therefore different platforms have been used, including tracked and multiple wheeled robots. The platforms are also dependent on the required task, for example, heavy crop harvesting (such as volume arable or root vegetables) will need a heavier platform than soft fruit picking. Legged robots have the potential of minimising their footprint, while maximising the flexibility of locomotion (e.g. moving sideways or in narrow spaces between crops, etc). Their agility, combined with the possibility of carrying specialised sensors, may help unlock the full potential of precision agriculture.
Manipulators will be needed for a range of tasks in future agriculture, replacing dexterous human labour, reducing costs and increasing quality, or performing operations more selectively than current larger machinery like slaughter harvesters. Work in this direction is ongoing, with soft grippers used for experimental work on selectively harvesting mushrooms, sweet peppers, tomatoes, raspberries and strawberries. Other applications such as broccoli harvesting can be performed with cutting tools, but will also require gentle handling and storage of the picked crop. In the open field, and for protected crops, there are complementary tasks to harvesting where manipulators can also play an important role. This includes mechanical weeding, precision spraying, and other forms of inspection and treatment. Manipulators will also be needed for the increased automation seen in food handling applications, such as large automated warehouses.
SENSING AND PERCEPTION
The integration of sensor systems within autonomous robotic systems offers the significant potential for new measurements that would otherwise be unobtainable. For example, current work addresses large area field mapping for bulk moisture by mobile robots, through the application of cosmic ray sensors adapted from the static COSMOS approaches . Significant advances in satellite- or drone-based remote sensing capabilities open opportunities in monitoring crop growth status with unprecedented temporal and spatial resolutions while at an affordable cost. Many open source satellite datasets (e.g. from the European Space Agency ) are freely available for farmers. Robotic platforms further offer the possibility of forensic testing of soil with the geotagging and immediate results from sampling sensors (such as laser-induced breakdown spectroscopy), or secure collection of samples for later analysis in a systematic and uncontaminated manner. The use of compact robots and on-board secure collection systems will further enable a step change in the regulatory efficiency and reliability of land management systems using robots.
Localisation and mapping
The use of GPS navigation in agriculture has become almost ubiquitous with the deployment of RTK (Real Time Kinematics) allowing accuracy of centimetres for the automated positioning of large farm machinery such as tractors and combined harvesters. More recently, approaches using data manipulation of the GPS signal alone have shown promise to deliver equivalent accuracy without the cost of extra radio beacons. Accurate location data is not confined to unmanned vehicles with GPS, as precise localisation systems are available using visual fiducial markers and/or optical, acoustic or radio beacons, depending on the speed and accuracy required. Sensor information is also required in detecting objects and risks in field in order to ensure safe operation of robotic vehicles. To minimise damage to crops, the accuracy of relative positioning and navigation is more important than that of absolute navigation and position as provided by RTK GPS in many applications. For example, it is desirable to drive robotic vehicles to follow crop lines in accuracy of centimetres or follow the tracks left by previous tractor operations. Multi-modal systems based on a combination of GPS, INS, LiDAR, vision, etc have further potential for providing accurate and robust solutions, without requiring in-field infrastructure such as beacons.
Several attempts have been made to utilise seed and weed mapping concepts by passively recording their geospatial location using RTK GPS. Farming robots can be further equipped with pattern classification techniques that can predict the density and species of different weeds using computer vision. Other methods focus on a dense semantic weed classification in multispectral images captured by UAVs .
With the addition of advanced vision systems, including depth perception, scanning sensors such as LiDAR and artificial intelligence for decision making and classification, the concept of precision can be taken to another level. The ability offered by ground based robots to precisely control the location of scanning sensors, such as LiDAR, opens up the possibility to return quantified biomass estimates over whole crops as well as related phenotypic data, such as growth rates and morphology, through the integration of accurate location data with rangefinder scans using simultaneous localisation and mapping (SLAM) techniques. Similarly, robotic sensing platforms offer the potential for broad area analysis of insect pest or pollinator movement and their speciation, utilising 3D microphones alone or in combination with light backscatter measurements to enable daylight measurements of characteristic flight trajectories. Thematic maps can be built up for diseases, pests or weeds, which enable variable rate treatments, a key concept in precision agriculture.
The use of both land-based and aerial platforms can allow the third dimension to be accurately added to the management of crops using data fusion and SLAM techniques. This can be combined with virtual reality or augmented reality (VR/AR) systems to provide monitoring and intervention possibilities to an individual plant scale. Long-term data collection will further enable the modelling of crops over time, for example, tracking the development of the crop canopy, and thus improved prediction of future growth patterns.
Such ground and aerial robotic platforms offer additional prospects for enabling localised extremely high signal-to-noise, high resolution sensing that may not be achieved by passive remote (satellite) or semi-remote (rotary or fixed wing UAV) sensing technologies. At the simplest level, these robotic platforms offer the potential to extract close proximity (within 10s of millimetres) reflectance and transmission.
Multispectral Imaging (MSI) data helps compensate for the erroneous measures that occur due to the surface topology and orientation of individual crop tissues. At a more advanced level, the use of robotic manipulators to locate sensors around crops or livestock could enable responses to be tested and examined, through applying artificial stimuli. For example, through applying a focused beam of light at specific areas of crop tissue, and then modulating the spectrum and intensity, it is possible to drive photochemistries within specific parts of plants, e.g. stems, young leaves, senesced older leaves, etc., which can then be sensed via multispectral imaging. In this way significantly greater phenotype information may be recovered from across plants than could be achieved by passive fixed imaging detectors alone. Similarly, the cell structures and arrangements within fruits, vegetable and meats may be non-destructively examined in high resolution, e.g. for mapping subcutaneous bruising in fruits or fat ratios in meats. Nutrient and water stress of crops can be assessed by fusing MSI data with other data sources. Combining these assessments with crop growth models gives a better prediction of yield and loss, which leads to improved farming management and better food supply chain management.
Machine vision approaches offer significant opportunities for enabling autonomy of robotic systems in food production. Vision-based tasks for crop monitoring include phenotyping , classifying when individual plants are ready for harvest , and quality analysis , e.g. detecting the onset of diseases, all with high throughput data. Vision systems are also required for detection, segmentation, classification and tracking of objects such as fruits, plants, livestock, people, etc., and semantic segmentation of crops versus weeds [32, 33, 34], etc. to enable scene analysis (understanding “what” is “where” and “when”) and safe operation of robotic systems in the field. Robotic vision in agriculture requires robustness to changes in illumination, weather conditions, image background and object appearance, e.g. as plants grow, while ensuring sufficient accuracy and real-time performance to support on-board decision making and vision-guided control of robotic systems. Active vision approaches, integrating next-best-view planning, may be needed to ensure that all the relevant information is available for robotic decision-making and control, e.g. where the fruit or harvestable part of a crop is occluded by leaves or weeds. Approaches based on analysis of 3D point clouds, e.g. derived from stereo imagery or RGB-D cameras, offer significant promise to achieve robust perception in challenging agricultural environments [30, 35].
Machine vision is already making an early impact in animal monitoring, e.g. for weight estimation, body condition monitoring  and illness detection  in pigs, cattle and poultry. Individual animal identification, e.g. using facial recognition techniques adapted from work in human facial biometrics , will allow more targeted precision care and timely interventions for individual animals, thereby ensuring their healthcare and wellbeing as well as optimising farm production.
Robotic vision often depends closely on machine learning from real-world datasets, with approaches such as deep neural networks [39, 29, 40] gaining traction and further raising the possibility for robots to share their knowledge by learning from Big Data. An open challenge in robotic vision and machine perception for robotic agriculture is to enable open-ended learning, facilitating adaptation to seasonal changes, new emerging diseases and pests, new crop varieties, etc. Most existing work considers only the initial training phase prior to deployment of a robot vision system, but not the ongoing adaptation of the learned models during long-term operation. The development of user interfaces for “ground truthing” and semi-supervised learning in robotic vision systems for agriculture is also an open challenge.
PLANNING AND COORDINATION
The true potential of robotics in agriculture will be harnessed when different types of robots and autonomous systems are brought together in a systemic approach. For example, UAVs are an excellent platform for environment monitoring, but with limited payloads and operational durability they are constrained when it comes to delivery of intervention or treatments on a larger scale. Hence, ground and airborne vehicles need to be integrated into heterogeneous fleets, coordinated either centrally or in a distributed fashion.
Planning, scheduling and coordination are fundamental to the control of multi-robot systems on the farm, and more generally for increasing the level of automation in agriculture and farming. For example, intelligent irrigation systems can respond to the change of weather conditions and crop growth status to automatically optimise the irrigation strategy so as to reduce the use of fresh water without loss of yields. The optimised strategy (e.g. when, where and the amount of water) is then implemented by computer-controlled irrigation equipment.
Such coordinated fleets also pose requirements for in-field communication infrastructure, such as Wi-Fi meshes, WiMAX ad-hoc networks, 5G approaches or other proprietary peer-to-peer communication methods deployed in field. On a larger scale, the heterogeneous fleets deployed in-field can also include collaborating humans sharing the working environment with their robotic counterparts, giving rise to interaction and communication requirements between the robots and the human operator and workers in this context. Example applications include in-field logistics, where vehicles need to be scheduled for area coverage and routing problems.
More generally, holistic approaches to fleet management are required, which fully integrate component methods for goal allocation, motion planning, coordination and control . These sub-problems have so far largely been studied in isolation, so basic research on integration and scaling to real-world scenarios is required. Aspects of swarm robotics could potentially be applied to fleet management systems in agriculture, as in the EU-funded ECHORD++ projects SAGA and MARS . To enable robot-human collaboration, the fleets also need to be aware of the presence of humans and to predict likely human actions in order to anticipate potential collisions and ensure safety.
In return, the motion of robotic systems needs to be legible to humans, to facilitate acceptance by and cooperation with their human counterparts.
Automated manipulation and grasping of food items presents a series of unique challenges compared to other sectors. These include significant natural size and shape variations between examples of the same product, heterogeneous positioning of products (e.g. during harvesting) and the fragile nature of food products. Some areas of food harvesting have been successfully automated but these solutions are best suited to situations where the entire content of a field becomes ready for harvest at the same time, e.g. grains or root vegetables. If plants fruit over an extended period of time with only some ready to harvest at any particular time (e.g. tomatoes) automation struggles. This is because discrete items must be harvested individually without disturbing those around them and, due to the dexterity, advanced perception and decision-making required, human labour is still widely used.
Soft robotics [43, 44] is expected to play an important role. Soft end-effectors and grippers are needed for gently handling soft fruit and vegetables, such as soft robot hands for lettuce harvesting and suction devices for picking apples. Robots are increasingly made softer also on the actuator/joint level. Whereas stiff robot arms are suitable for blind operation in a factory environment, an agricultural manipulator requires sensorimotor coordination to achieve its task. Some tasks also require the right amount of force to be applied, dictating a force-based rather than position-based approach to control. In general, grasping and manipulation applications in Agri-Food require robustness to the unforeseen, while maintaining their ability to actuate with precision. One way to achieve this is through variable-stiffness actuators , which incorporate elastic structures, much like humans.
The development of compliant manipulators and grippers will in turn transform and simplify the design of agricultural robots by reducing the need for complex visual and tactile sensors. For this potential to be fully unlocked, novel design and control techniques need to be developed. Grasp planning is also a significant challenge. The most common approach is to use vision systems to locate products and use this to direct the grasp. However, this approach can fail if the object to be grasped is partially obscured by other products or foliage. Vision alone provides only limited data about an object during grasping and picking; human operators also use tactile feedback to adjust their action as a product is grasped to ensure it is picked successfully.
The challenges for interaction range from domain-independent aspects such as intuitive designs, immersive displays (e.g. Virtual and Augmented Reality) and tactile feedback, to very specific challenges stemming from the in-field conditions. Examples include the design of suitable interaction devices that are operable under harsh conditions, with constrained dexterity and precision of the operators, e.g. workers wearing gloves or having wet and muddy hands, or to guarantee the safety of often large and heavy semi-autonomous machinery in an environment shared with human workers. In contrast to robots in factories, where working areas can be fenced off when a robot is in operation, agricultural robots are limited by the absence of safety infrastructure in the fields, and require new innovative solutions.
Robots closely collaborating with humans (so-called cobots ) are delivering real step changes in many industrial sectors, and are anticipated to be vital to automation in agriculture. Use cases range from farm in-field logistics (transportation), where efficient and safe hand-over of goods and produce needs to be facilitated, to applications enhancing animal and crop welfare by means of integrated monitoring and intervention delivery. An illustrative example is the RASberry project at the University of Lincoln, where human pickers of strawberries are supported by mobile robots acting as transporters.
While some tasks for cobots require physical interaction between robots and humans, in other areas robots can act as a mediator or provide a remote presence for agronomists and farmers. Therefore a focus on intuitive and ergonomically appropriate interfaces and interaction design is needed. Concepts of shared autonomy and control, allowing operators to exercise control from remote locations over a potentially heterogeneous (ground, air, water) fleet of semi-autonomous robots, will also be important. As the technology matures, and in particular for safety-critical tasks, various levels of shared autonomy will be seen, where the human operator guides the high-level execution, while the robotic system performs the required sensorimotor coordination on the ground. The fan-out , or number of robots a human can control simultaneously, will help drive the mixture of human supervisors and robot agents in such a paradigm.
Safe human-robot interaction
By relying on humans as supervisors, the autonomy levels, and associated risks and design complexities, can also be improved. Human supervision will be a vital safety factor for most agri-robotic systems for the foreseeable future, while the technology develops towards higher levels of autonomy. The robotic systems will also be learning and adapting to task and farm-specific constraints. Human and robot collaborators will therefore likely be mutually adapting to each other, in order to maximise performance.
Approaches to safe physical Human Robot Interaction (pHRI) [48, 49] include supervisory systems to monitor the interaction and adjust the behaviour of the robot if an unsafe situation is identified. This typically involves slowing, or completely stopping the robot, to prevent accidents. However, this approach can significantly reduce productivity as the robot is not working to its full potential. Current research aims to improve on this approach by allowing robots to identify and predict unsafe situations, and then to adapt and adjust their operation to continue the task in a manner that allows both productivity and safety to be maintained . A further approach to ensuring safe pHRI is to design robot systems which are inherently safe, meaning that if collisions occur between human and machine, injury will not result. The aim is to replicate the safe interaction that occurs when multiple people work collaboratively.
This requires a change away from heavy, rigid and high inertia robots to systems which are more akin to biological creatures. Again this is a challenge that the new field of soft robotics may be able to address.
LEARNING AND ADAPTATION
Artificial intelligence technologies, especially in machine learning, are expected to play a major role in most of the above technology areas, and will be essential enablers for agricultural robots. Agricultural environments are subject to changes throughout the lifetime of a robotic system. For example, there may be new crop varieties, weeds, pests, diseases, treatments, legislation, climate change, etc., as well as new implements and robotic technologies. In AI terms this means dealing with an open world, so techniques to enable adaptation during operation rather than at the design phase will be crucial. Techniques that allow robots to learn from experience include reinforcement learning, learning from demonstration, and transfer learning to exploit prior knowledge, e.g. from another domain or task. Ongoing research is investigating deep learning methods , especially in perception-related tasks involving the interpretation of sensor data, including recognition and segmentation tasks in automated weeding and fruit picking. Robots will also need to leverage human knowledge, especially when facing situations that were not foreseen at design time. This additional input might be given by end-users, maintainers, and/or domain experts. It might also be provided through direct control (i.e. teleoperation), natural interaction (e.g. via language or gestures) or by the means of labelled examples and data sets. These developments will link naturally into the use of Big Data in smart farming , alongside the use of satellite imaging, UAVs and ground robots for more localised and richer, multimodal data collection. These developments coupled with cloud-based storage will create an abundance of information that could potentially be utilised for smart planning and control of agriculture. An important requirement is the standardisation of data to ease the exchange between robots, domains, farms, countries and companies.
Research efforts for development of agricultural robots that can effectively perform tedious field tasks have grown significantly in the past decade. With the exception of milking robots that were invented in the Netherlands, robotics has not reached a commercial scale for agricultural applications. With the decrease of the workforce and the increase of production cost, research areas on robotic weeding and harvesting have received more and more attention in the recent years, however the fastest available prototype robots for weeding and harvesting are not even close to being able to compete with the human operator. For the case of picking valuable fruits using robots, the technology is now becoming closer to a commercial product with the emerging of the SWEEPER. For other fruits such as citrus and apples that can be mass harvested for juice industry, modifications of the existing mechanical harvesting systems with some robot functionalities may be more promising than using single robot system. Increasing the speed and accuracy of robots for farming applications are the main issues to be addressed for generalization of robotics systems, however, compared to the industrial and military cases, the lack of abundant research funding and budgets in agriculture has decelerated this process. For the case of robot harvesting, improving sensing (fruit detection), acting (manipulator movement, fruit attachment, detaching, and collecting), and growing system (leave pruning and plant reshaping) are suggested to increase the efficiency. It should be noted that development of an affordable and effective agriculture robot requires a multidisciplinary collaboration in several areas such as horticultural engineering, computer science, mechatronics, dynamic control, deep learning and intelligent systems, sensors and instrumentation, software design, system integration, and crop management. We highlighted some of the facing challenges in the context of utilizing sensors and robotics for precision agriculture and digital farming as: object identification, task planning algorithms, digitalization, and optimization of sensors. It was also mentioned that for an autonomous framework to successfully execute farming tasks, research focus should be toward developing simple manipulators and multi-robot systems. This is in fact one of the academic trends and research focuses in agricultural robotics for building a swarm of small-scale robots and drones that collaborate together to optimize farming inputs and reveal denied or concealed information. As of the conclusion, some forms of human-robot collaboration as well as modification of the crop breeding and planting systems in fields and greenhouses might be necessary to solve the challenges of agricultural robotics that cannot yet be automated. For example, in a collaborative harvesting system using human-and-robot, any fruit that is missed by the robot vision will be spotted by the human on a touchscreen interface. Alternatively, the entire robot sensing and acting mechanism can be performed by a human operator in a virtual environment. Nevertheless, an agricultural robot must be economically viable which means it must sense fast, calculate fast, and act fast to respond to the variability of the environment.