A brief discussion on the main market and driving sources of service robots

Entertainment and Education

This sector has achieved more transformational robotics than any other, especially in addressing the nation’s science, technology, engineering, and math (STEM) crisis while becoming a true “4R” education. FIT’s huge success is a testament to this. Robotics provides an interesting and accessible way for children to learn and apply the basics of math and science, including engineering and system integration principles, to produce intelligent machines to accomplish specific tasks.

Factors affecting commercialization

If all of the above areas are realized, then a lot of investment will be needed to expand and develop robotics. As mentioned above, there is still a long way to go before fully autonomous robotics, that is, robotics that can operate automatically without human instructions or intervention. The scholars at the conference reached a consensus that the progress of robotics technology has made it possible to develop and commercialize the basic and application of robots, and can significantly "enhance human functions."

These solutions will be able to automatically adjust based on the following functions: monitor the dynamic physical environment in a deterministic manner, identify targets, detect changes, sense environmental conditions, analyze and respond based on detected situations, respond based on human commands and execute actions automatically within pre-authorized boundaries without operator intervention.

Examples of these robotic solutions include tele-surgery, such as the da Vinci surgical system, and autonomous professional robots, such as Roomba. As the Internet continues to grow, the natural progression will be from remote sensing to remote manipulation. This expansion of the Internet into the physical world will help further blur the lines between communications, computing, and services, inspiring applications of remote communication and remote participation. More realistic solutions will emerge that have distributed cognitive capabilities and can effectively leverage human intelligence. These solutions will be combined with robotics to enable autonomous location sensing while allowing operators to intervene as needed from a distance via the Internet.

Based on the above, the aging population will lead to a future labor shortage. As workers seek to move to higher levels of their careers, the automation of low-level jobs will increase, and the number of workers doing low-level jobs will gradually decrease or even disappear. The challenge of achieving fully automated solutions in the long term will continue to exist due to technological limitations, and the short-term challenge is to investigate the needs of its development and decide how to best "cross the gap". That is, identifying the right value proposition, cost reduction, effective development, effective system engineering process, deciding how to best integrate solutions, and how to transform technology into products.

Scientific and technological challenges

Mobility

Mobility is a paradigm shift in robotics research. This success is reflected in many systems that have demonstrated performance in real-world environments, including museum tour guides and autonomous vehicles in the DARPA Motorcycle Challenge and City Challenge. However, the panelists agreed that a number of important questions remain unanswered. Finding answers to these questions in the area of ​​mobility will be critical to achieving autonomous control and versatility in related areas of robotics.

Three-dimensional navigation is one of the most important challenges in the mobility space. Currently, most mapping, localization, and navigation systems rely on flat representations of the Earth, such as street maps for terrestrial tasks. However, as the complexity of robotic applications increases, and new robots are deployed every day, these two-dimensional representations are insufficient to capture the necessary context in unmodeled, uncontrolled, crowded environments. Therefore, it will be important to acquire three-dimensional world models that support navigation and manipulation. These three-dimensional representations should not include the geometric layout of the world; instead, the map must contain task-relevant semantic information about the objects in the environment and their features.

Currently, robots have a good understanding of where objects are in the physical world, but little or no understanding of what objects are. When it comes to grasping and the services of environmental representation to perform mobility functions, environmental representation should also include object context support (i.e., information about what the robot can do with an object). Achieving semantic 3D navigation will require new methods for sensing, perception, map matching, localization, object recognition, context support recognition, and planning.

3D mapping technology is the construction of maps using different kinds of sensors. Currently, robots rely on laser measurement systems or game control distance sensors such as Microsoft's Kinect or PrimeSense to obtain environmental information, using a mapping called "SLAM". Some experts have proposed that the field of "visual SLAM" (VSLAM) should be further developed away from laser measurement systems. This technology relies on cameras (robust, cheap, and easily available sensors) for mapping and positioning in the 3D world. Currently, VSLAM systems have demonstrated impressive performance. Therefore, it is believed that VSLAM may play an important role in the development of 3D navigation capabilities with sufficient information and affordable prices.

The additional requirement of 3D navigation for a specific application, outdoor 3D navigation, also presents a number of important challenges that need to be explicitly addressed. Among these challenges is the fact that current 2D environmental representations cannot capture the complexity of the outdoor environment, nor the outdoor light conditions, which are factors that cause sensor performance variations. Also, how to achieve navigation in crowds is an important challenge.

operate

Almost all service robots need substantial progress in operational performance. These applications require the robot to physically interact with the environment, including opening doors, picking up objects, operating machines and equipment, etc. At present, autonomous operating systems work well in precisely engineered and highly controlled environments, such as factory assembly cells, but are unable to cope with changes and uncertainties in open, dynamic and unmodeled environments. Therefore, scholars from the three frontier discussion groups believe that "autonomous operation" is its key area. Although no specific research progress direction has been determined, the scholars' discussions revealed that the basic assumptions of most existing operation algorithms cannot be met in practical applications. Whether it is possible or not, grasping and operation suitable for open, dynamic, and unstructured applications should utilize prior knowledge and environmental models. In the absence of prior knowledge, catastrophic consequences should not result. As a corollary, when the environmental model does not exist, true autonomous operation will rely on the robot's ability to obtain sufficient, task-related environmental models. Compared with most existing methods that emphasize planning and control, this means that perception will become an important research issue in the autonomous operation research agenda.

Pick and place operations can provide a sufficient functional foundation for many well-defined application operational requirements. Therefore, pick and place operations of increasing complexity and versatility can provide a route and benchmark for research efforts in autonomous manipulation.

planning

Research in the field of motion planning has made great progress in the past decade, with algorithms and techniques impacting many different application areas. However, robust, dynamic 3D path planning remains an unsolved problem. An important factor in this problem is the concept of robot position awareness (i.e., the robot can autonomously integrate, intersect, and integrate behavior planning using "appropriate" sensing and modeling methods). "Appropriate" means that a complete and accurate model of the environment is not available to the robot in real time. Instead, it is necessary to make inferences about objects, the environment, perception, and robot behavior. This leads to a blurring of the line between planning and motion planning. To plan a motion, the planner needs to coordinate sensing and motion with constraints imposed by the task. To achieve the task goals robustly and reliably, planning needs to take into account the support of the task environment. This means that the planner needs to consider the environment and the interactions between objects in the environment as part of the planning process.

For example, to pick up an object, it may be necessary to open a door, enter a different room, push a chair to reach a cupboard, open a cupboard door, and push an obstacle. Within this new planning paradigm, the task and the constraints imposed by the task and the environment are key; the "motion" in "motion planning" is a way to reach the end point. Planning under the constraints considered in the planning process comes from target grasping, motion (such as step planning), kinematics and dynamics of the mechanism, posture constraints, and obstacle avoidance. Planning under these constraints requires the robot system to be real-time.

The motion of a robot can easily be constrained by the feedback from sensors. The most obvious examples are contact constraints and obstacle avoidance. Therefore, feedback planning and the integration of control and planning are important research topics to meet the planning needs raised by the scholars at the conference. The feedback planner generates a strategy that directly maps states to behaviors, rather than generating specific paths or trajectories. This ensures that the uncertainties of sensors, actuators, and models can be resolved through sensor feedback.