Designing a logistic network or system is a complex task. It involves a great deal of understanding of the business, a clear vision of the demand, and a deep comprehension of related technologies. This report focuses on robotics technologies that are useful for logistic systems.
Due to the nature of logistics, favorable technologies usually have one or several of the following features: improving efficiency, lowering cost, reducing error rate, agile to changes, safe, fast deployment, a low impact on the existing layout, low integration overhead, etc. Especially for e-commerce, logistics systems would need to handle millions of goods in varies shapes, sizes, and weights and they are packaged differently. Robotics with AI and computer vision is often seen as the technology that holds solutions to such demands.
Many major logistics companies or e-commerce companies have invested big in robotics technologies for their logistics networks. Leaders such as Amazon, Alibaba (Cainiao Networks), UPS, FedEx, and DHL all have a significant robotics division working on robotics technologies and keep rolling out inspiring new robots. Recently, a number of robotics startups such as Fetch and Nuro set their focus on logistics because of its potential economic impact and market size. Even traditional robotics companies such as Hitachi, KUKA (Swisslog), FANUC have shifted their business focuses and lean toward logistics.
Among all robotics technologies, AGV is the most matured and widely used in logistics. This report starts with introducing the current and future of AGVs, and then home delivery robots, door-to-door autonomous vehicles, autonomous air freight transport, UAVs, palletizing robots, picking robots, and in-vehicle sorting robots. This report will end with an outlook into the robotics technology trend.
Automated Guided Vehicle (AGV)
AGVs started transporting material since the 1950s. The technology became mature in the 1980s. In recent years, AGV technology grows rapidly with the logistics industry. The logistics equipment market has been increasing by about 30% every year. In the last few years, two third of all AGVs sold in China are deployed in the logistics industry.
In many logistics scenarios, solutions based on AGVs have clear advantages over conveyors that have been used in manufacturing plants for more than 100 years since 1913. AGVs gain popularity because they fit well in today’s lean and flexible world. They move goods from point A to point B with a flexibility that allows both the moving route and key points to change without any infrastructure change. They can make an existing logistics system adapt to new business rules fast without a steep hardware cost. They do not suffer from single-point-of-failure because of their exchangeability. They can work in small and large scales, and handle various loads. So its initial investment can be small, and it is easy to scale up along with the business. They are more green in term of power efficiency since they consume power as needed while conveyors consume power constantly. But, conveyors are usually better than AGVs in stable warehouses where there are high and fixed goods-moving rates.
An AGV system is usually composed of AGVs, landmarks, charging stations, a communication system, a job control system, and a traffic management system. The job control system and traffic management system are usually centralized. Recently AGVs with localization and obstacle avoidance capabilities can plan their own path locally based on their own sensor inputs. Those AGVs may also have local coordination strategies. However, the local coordination strategies are usually limited and their individual path planners have to work with the centralized global traffic management and coordination strategies to ensure the overall efficiency of the warehouse.
A typical AGV would have a motor system, sensor modules, a controller, a communication module, a battery, and a payload interface. More advance AGVs may have an AI module. The sensor modules on a basic AGV usually provide two functionalities: localization and safety. The sensors for localization work with guides embedded in the infrastructure. Both the sensors and guides are designed at the same time.
Currently, AGVs are positioned using LASER reflectors, magnetic tapes, invisible UV markers, visual landmarks, ultra-wideband, or QR code tags as guides. Among all guided localization solutions, the QR AGVs are very popular since they have low setup costs, matured hardware and controllers while providing good performance. A typical QR AGV would have several mm precision with a speed of a few meters per second. Ultra-wideband is another new approach that gains more attention recently for indoor positioning. It is a short-range radio technology. Different from Bluetooth and Wi-Fi, Ultra-wideband positioning is done with the time of flight (ToF) instead of signal strengths. The transmitter and receiver have to have a direct line of sight. It has relatively low latency (< 10 ms), but poor accuracy (10-30 cm). So it won’t work in an environment with tight tolerance. But it works fine for applications having sparse AGVs in open and large spaces. For higher performance and in an environment with tight tolerance, LASER AGVs are more preferable.
Sensors for guided localization solutions do not really perceive the environment. The sensors are tailored to only perceive the guides. They cannot see obstacles. For safety, an AGV could be equipped with a safety bumper and sonar sensors to detect a closeup obstacle and avoid collision by making an emergency stop or taking a shape turn. But they cannot see far enough for planning a new smooth path to avoid the obstacle from a distance. More advanced sensors such as vision sensors and 2D or 3D Lidar would be needed.
Advanced vision and Lidar sensors can see not only obstacles from a distance but also natural landmarks. They would remove the necessity of installing any artificial landmarks or tags on the infrastructure. The AGVs equipped with vision and Lidar sensors could recognize natural landmarks, track them, and build a map based on the natural landmarks. They compute their positions by observing and identify the landmarks on the map, recognize obstacles, and plan paths to navigate among them freely. The process is called visual based simultaneous localization and mapping (vSLAM). A significant amount of work on this top has been carried out in robotics and computer vision communities. We call AGVs that can self-localize and plan paths based on their local information “autonomous AGVs.” Since warehouse environments are semi-structured, the current vSLAM techniques are sufficient to support applications using autonomous AGVs.
Other AGVs or other robotic vehicles are not considered as obstacles since they can communicate and collaborate. However, workers and manually operated vehicles (MOVs) such as forklifts are dynamic obstacles and they do not communicate with AGVs. Therefore it is a challenge to have the AGVs to collaborate with workers and MOVs. AGVs can see workers and MOVs either through their perception systems or receiving position readings from the positioning sensors such as ultra-wideband transmitters worn by the workers of embedded in MOVs. A moving geofence can be defined around a worker and the AGVs are programmed to never penetrate the geofence to ensure safety.
To efficiently and safely work with the workers, autonomous AGVs should not only perceive where the workers are but also predict where they would move to in the next few seconds. To do that, the AGVs should understand each individual worker’s goals and behavior patterns. The worker’s goals are usually associated with the assigned task. The worker’s behavior patterns could be learned from the worker’s history data. On the other side, the workers would need to know and understand the AGV’s behaviors to predict where the AGV would move to next. For this purpose, research works have been carried out to display AGV’s moving intention on AR goggles, or on the floor around the AGV using spatial augmented reality.
Overall, the guided AGV technology for fully structured warehouses is mature enough for most logistics applications. Because guided AGVs cannot work with moving human workers, they have to have reached to a significant number to completely take over the transportation job of a semi-closed area. Therefore, even though it is less costly and more flexible and scalable than conveyors, setting up a guided AGV system in the warehouse still requires a significant capital investment, a long setup and fine-tuning time. On the other hand, since autonomous AGVs can work with human workers, they can share transportation job load with human workers dynamically. Warehouses can adapt autonomous AGVs gradually without significant capital investment or long setup time. Comparing with autonomous cars, the autonomous AGVs work in a relatively simple and structured environment, they can achieve a satisfying safety standard with a weaker perception capability.