The state of industrial robotics: challenges & opportunities

Improvements to move industrial robotics forward

Based on what MIT learned about how the companies it spoke to have experimented with and integrated industrial robots and enabling technologies, it recommended improvements in the following eight areas.

Perception: These technologies pose a key limitation, both in the robustness of existing systems and in their use in downstream subsystems, such as grasping. The study said researchers should focus on improving the robustness of perception capabilities in factory environments, particularly in migrating recent advances, such as deep learning techniques, to be more effective. Such migrations require methods to scale appropriately, as vast amounts of data are unlikely to be available in such domains, and to ensure robustness to environmental variations such as changes in lighting.

Intelligent gripping: Pick and place is an essential robotics task. However, maneuvering objects in less constrained factory environments or allowing variations in the size and pose of a part, require a high degree of robotic intelligence. Fully automated, generalized bin-picking systems are an immediate area of focus for many companies and a partially solved problem, according to the study.

For the system to work efficiently, robot vision and sensory perception need to be advanced enough for robots to process objects from a pile of randomly distributed or scattered objects and pick them up. Research into dexterous in-hand manipulation, and improving technologies for tactile sensing and bin picking, including gripper design and grasp planning, in the context of a factory environment is an important research direction, as these technologies are critical to more flexible operations. Along with perception challenges, gripping remains one of the most limiting factors of automation in factories today, the study said.

Safe, collaborative robots: There are several areas of improvement here, according to the study, including integration of human behavior prediction into robot path planning, for better safety and collaboration; improved tracking in dynamic workspaces; as well as demonstrating, through viable working models, the advantages of collaborative workcells and robot assistance to human-led operations. Again, making sure that the entire robotic system, including the tasks that the human performs, is safe to work with should be a priority.

Autonomous guided vehicles: An AGV can use a navigation technique called simultaneous localization and mapping (SLAM) in lieu of traditional navigation techniques to localize itself within a space. However, the application of the technology in industry is still emerging. While SLAM provides robust or versatile data association, it is still hard to quantify how SLAM can improve logistical operation. Standards for fleet management are still emerging and are necessary to fully realize the benefits of AGVs for logistics. In addition, current SLAM capabilities need to be enhanced for the technology to be sufficient for mobile manipulation.

Interfaces and programming: Programming robots is a difficult and expensive task, requiring time and expertise. Research into human-robot interaction and methods to simplify the teaching or programming of robots, such as programming by demonstration and other methods not conditional on previous or significant programming expertise, are extremely valuable, especially those that allow the robot to generalize its behavior. These can ease skill requirements and, in turn, lower integration costs, resulting in wider use and application of robotics.

Simulation: More work needs to be done to enhance digital simulation techniques, such as digital twins, to include, for instance, more detailed and fine-grained simulations of workcells or manufacturing lines. While virtual reality and augmented reality are promising avenues of research, factories could more immediately benefit from more high-fidelity models that typically yield better prototypes.

Worker-centered design: To back up anecdotal and experiential evidence, more research is needed into how design and integration processes are affected by contributions of workers, and how this affects performance and overall production cycles. In the long run, this can create a working framework for governing and evaluating human-machine interactions in industrial settings.

Cloud systems: Cloud systems promise to connect and control operations on the factory floor through the use of sensor data. However, so far, this vision has not been realized. Wide-scale sensor integration incurs heavy costs and a possible overhaul of company infrastructure. At this point, many companies say that it is difficult to make sense of the data that they collect from factory operations, and that they do not know how to extract and extrapolate useful information from it. Work toward how to extract meaningful information from data is therefore a valuable next step.